From c5d109ee8c322ae94bd0aed60a48cb3d200523da Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Thu, 21 Oct 2021 19:52:50 -0300
Subject: [PATCH 01/28] Expose n_iter_ from libSVM and remove trailing
 whitespaces (pre-commit)

---
 sklearn/svm/_base.py                          |   9 +
 sklearn/svm/_classes.py                       |  33 +++
 sklearn/svm/_libsvm.pxi                       |   1 +
 sklearn/svm/_libsvm.pyx                       |  14 +-
 sklearn/svm/_libsvm_sparse.pyx                |  11 +-
 sklearn/svm/src/libsvm/LIBSVM_CHANGES         |   1 +
 sklearn/svm/src/libsvm/libsvm_helper.c        |  11 +
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c |  11 +
 sklearn/svm/src/libsvm/svm.cpp                | 210 ++++++++++--------
 sklearn/svm/src/libsvm/svm.h                  |   2 +
 10 files changed, 205 insertions(+), 98 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 9c992e0d1f1fa..1522edee345c0 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -266,6 +266,13 @@ def fit(self, X, y, sample_weight=None):
             self.intercept_ *= -1
             self.dual_coef_ = -self.dual_coef_
 
+        # If the number of models optimized by libSVM is one, get the number of
+        # iterations as an integer instead of ndarray.
+        if (
+            self._impl in ["c_svc", "nu_svc"] and len(self.classes_) <= 2
+        ) or self._impl in ["one_class", "epsilon_svr", "nu_svr"]:
+            self.n_iter_ = self.n_iter_[0]
+
         return self
 
     def _validate_targets(self, y):
@@ -312,6 +319,7 @@ def _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
             self._probA,
             self._probB,
             self.fit_status_,
+            self.n_iter_,
         ) = libsvm.fit(
             X,
             y,
@@ -352,6 +360,7 @@ def _sparse_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
             self._probA,
             self._probB,
             self.fit_status_,
+            self.n_iter_,
         ) = libsvm_sparse.libsvm_sparse_train(
             X.shape[1],
             X.data,
diff --git a/sklearn/svm/_classes.py b/sklearn/svm/_classes.py
index 719540cd725c0..0172fe6c6e983 100644
--- a/sklearn/svm/_classes.py
+++ b/sklearn/svm/_classes.py
@@ -670,6 +670,15 @@ class SVC(BaseSVC):
 
         .. versionadded:: 1.0
 
+    n_iter_ : int or ndarray of shape (n_class*(n_class-1)/2,)
+        Number of iterations run in optimization. If `classes_ <= 2`, only one
+        model is optimized, thus an integer is returned. Otherwise, multiple
+        models are optimized separately, thus having multiple number of
+        iterations. In this case a numpy array is returned with the number of
+        iterations for each model.
+
+        .. versionadded:: 1.1
+
     support_ : ndarray of shape (n_SV)
         Indices of support vectors.
 
@@ -925,6 +934,15 @@ class NuSVC(BaseSVC):
 
         .. versionadded:: 1.0
 
+    n_iter_ : int or ndarray of shape (n_class*(n_class-1)/2,)
+        Number of iterations run in optimization. If `classes_ <= 2`, only one
+        model is optimized, thus an integer is returned. Otherwise, multiple
+        models are optimized separately, thus having multiple number of
+        iterations. In this case a numpy array is returned with the number of
+        iterations for each model.
+
+        .. versionadded:: 1.1
+
     support_ : ndarray of shape (n_SV,)
         Indices of support vectors.
 
@@ -1140,6 +1158,11 @@ class SVR(RegressorMixin, BaseLibSVM):
 
         .. versionadded:: 1.0
 
+    n_iter_ : int
+        Number of iterations run in optimization.
+
+        .. versionadded:: 1.1
+
     n_support_ : ndarray of shape (n_classes,), dtype=int32
         Number of support vectors for each class.
 
@@ -1328,6 +1351,11 @@ class NuSVR(RegressorMixin, BaseLibSVM):
 
         .. versionadded:: 1.0
 
+    n_iter_ : int
+        Number of iterations run in optimization.
+
+        .. versionadded:: 1.1
+
     n_support_ : ndarray of shape (n_classes,), dtype=int32
         Number of support vectors for each class.
 
@@ -1512,6 +1540,11 @@ class OneClassSVM(OutlierMixin, BaseLibSVM):
 
         .. versionadded:: 1.0
 
+    n_iter_ : int
+        Number of iterations run in optimization.
+
+        .. versionadded:: 1.1
+
     n_support_ : ndarray of shape (n_classes,), dtype=int32
         Number of support vectors for each class.
 
diff --git a/sklearn/svm/_libsvm.pxi b/sklearn/svm/_libsvm.pxi
index ab7d212e3ba1a..46b85f89a0738 100644
--- a/sklearn/svm/_libsvm.pxi
+++ b/sklearn/svm/_libsvm.pxi
@@ -56,6 +56,7 @@ cdef extern from "libsvm_helper.c":
                          char *, char *, char *, char *)
 
     void copy_sv_coef   (char *, svm_model *)
+    void copy_num_iter  (char *, svm_model *)
     void copy_intercept (char *, svm_model *, np.npy_intp *)
     void copy_SV        (char *, svm_model *, np.npy_intp *)
     int copy_support (char *data, svm_model *model)
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index 9488bda4ccf58..921fb03373e3b 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -155,6 +155,10 @@ def fit(
 
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates, empty array for probability=False.
+
+    n_iter_ : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
+              else (1,)
+        Number of iterations run in optimization.
     """
 
     cdef svm_parameter param
@@ -202,6 +206,14 @@ def fit(
     SV_len  = get_l(model)
     n_class = get_nr(model)
 
+    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter
+    if n_class > 2:
+        num_iter = np.empty (int((n_class*(n_class-1))/2), dtype=np.int32)
+        copy_num_iter (num_iter.data, model)
+    else:
+        num_iter = np.empty (1, dtype=np.int32)
+        copy_num_iter (num_iter.data, model)
+
     cdef np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef
     sv_coef = np.empty((n_class-1, SV_len), dtype=np.float64)
     copy_sv_coef (sv_coef.data, model)
@@ -251,7 +263,7 @@ def fit(
     free(problem.x)
 
     return (support, support_vectors, n_class_SV, sv_coef, intercept,
-           probA, probB, fit_status)
+           probA, probB, fit_status, num_iter)
 
 
 cdef void set_predict_params(
diff --git a/sklearn/svm/_libsvm_sparse.pyx b/sklearn/svm/_libsvm_sparse.pyx
index 92d94b0c685a5..3228aa7d7f766 100644
--- a/sklearn/svm/_libsvm_sparse.pyx
+++ b/sklearn/svm/_libsvm_sparse.pyx
@@ -41,6 +41,7 @@ cdef extern from "libsvm_sparse_helper.c":
                                   double, int, int, int, char *, char *, int,
                                   int)
     void copy_sv_coef   (char *, svm_csr_model *)
+    void copy_num_iter  (char *, svm_csr_model *)
     void copy_support   (char *, svm_csr_model *)
     void copy_intercept (char *, svm_csr_model *, np.npy_intp *)
     int copy_predict (char *, svm_csr_model *, np.npy_intp *, char *, BlasFunctions *)
@@ -159,6 +160,14 @@ def libsvm_sparse_train ( int n_features,
     cdef np.npy_intp SV_len = get_l(model)
     cdef np.npy_intp n_class = get_nr(model)
 
+    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter
+    if n_class > 2:
+        num_iter = np.empty (int((n_class*(n_class-1))/2), dtype=np.int32)
+        copy_num_iter (num_iter.data, model)
+    else:
+        num_iter = np.empty (1, dtype=np.int32)
+        copy_num_iter (num_iter.data, model)
+
     # copy model.sv_coef
     # we create a new array instead of resizing, otherwise
     # it would not erase previous information
@@ -217,7 +226,7 @@ def libsvm_sparse_train ( int n_features,
     free_param(param)
 
     return (support, support_vectors_, sv_coef_data, intercept, n_class_SV,
-            probA, probB, fit_status)
+            probA, probB, fit_status, num_iter)
 
 
 def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
diff --git a/sklearn/svm/src/libsvm/LIBSVM_CHANGES b/sklearn/svm/src/libsvm/LIBSVM_CHANGES
index bde6beaca2694..7d594c826e5f5 100644
--- a/sklearn/svm/src/libsvm/LIBSVM_CHANGES
+++ b/sklearn/svm/src/libsvm/LIBSVM_CHANGES
@@ -7,4 +7,5 @@ This is here mainly as checklist for incorporation of new versions of libsvm.
   * Improved random number generator (fix on windows, enhancement on other
     platforms). See <https://github.com/scikit-learn/scikit-learn/pull/13511#issuecomment-481729756>
   * invoke scipy blas api for svm kernel function to improve performance with speedup rate of 1.5X to 2X for dense data only. See <https://github.com/scikit-learn/scikit-learn/pull/16530>
+  * Expose the number of iterations run in optimization.
 The changes made with respect to upstream are detailed in the heading of svm.cpp
diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index 17f328f9e7c4c..bfb150235d30b 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -218,6 +218,16 @@ npy_intp get_nr(struct svm_model *model)
     return (npy_intp) model->nr_class;
 }
 
+/*
+ * Get the number of iterations run in optimization
+ */
+void copy_num_iter(char *data, struct svm_model *model)
+{
+    int n_models;
+    n_models = model->nr_class <= 2 ? 1 : model->nr_class * (model->nr_class-1)/2;
+    memcpy(data, model->num_iter, n_models * sizeof(int));
+}
+
 /*
  * Some helpers to convert from libsvm sparse data structures
  * model->sv_coef is a double **, whereas data is just a double *,
@@ -371,6 +381,7 @@ int free_model(struct svm_model *model)
     free(model->label);
     free(model->probA);
     free(model->probB);
+    free(model->num_iter);
     free(model->nSV);
     free(model);
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index a85a532319d88..7234e029b32fd 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -348,6 +348,16 @@ void copy_sv_coef(char *data, struct svm_csr_model *model)
     }
 }
 
+/*
+ * Get the number of iterations run in optimization
+ */
+void copy_num_iter(char *data, struct svm_csr_model *model)
+{
+    int n_models;
+    n_models = model->nr_class <= 2 ? 1 : model->nr_class * (model->nr_class-1)/2;
+    memcpy(data, model->num_iter, n_models * sizeof(int));
+}
+
 /*
  * Get the number of support vectors in a model.
  */
@@ -409,6 +419,7 @@ int free_model(struct svm_csr_model *model)
     free(model->label);
     free(model->probA);
     free(model->probB);
+    free(model->num_iter);
     free(model->nSV);
     free(model);
 
diff --git a/sklearn/svm/src/libsvm/svm.cpp b/sklearn/svm/src/libsvm/svm.cpp
index d209e35fc0a35..84e9ca09c44a2 100644
--- a/sklearn/svm/src/libsvm/svm.cpp
+++ b/sklearn/svm/src/libsvm/svm.cpp
@@ -31,7 +31,7 @@ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-/* 
+/*
    Modified 2010:
 
    - Support for dense data by Ming-Fang Weng
@@ -55,6 +55,9 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
      Sylvain Marie, Schneider Electric
      see <https://github.com/scikit-learn/scikit-learn/pull/13511#issuecomment-481729756>
 
+   Modified 2021:
+
+   - Exposed number of iterations run in optimization, Juan Martín Loyola.
  */
 
 #include <math.h>
@@ -125,7 +128,7 @@ static void info(const char *fmt,...)
 and dense versions of this library */
 #ifdef _DENSE_REP
   #ifdef PREFIX
-    #undef PREFIX  
+    #undef PREFIX
   #endif
   #ifdef NAMESPACE
     #undef NAMESPACE
@@ -136,7 +139,7 @@ and dense versions of this library */
 #else
   /* sparse representation */
   #ifdef PREFIX
-    #undef PREFIX  
+    #undef PREFIX
   #endif
   #ifdef NAMESPACE
     #undef NAMESPACE
@@ -163,7 +166,7 @@ class Cache
 	// return some position p where [p,len) need to be filled
 	// (p >= len if nothing needs to be filled)
 	int get_data(const int index, Qfloat **data, int len);
-	void swap_index(int i, int j);	
+	void swap_index(int i, int j);
 private:
 	int l;
 	long int size;
@@ -439,7 +442,7 @@ double Kernel::dot(const PREFIX(node) *px, const PREFIX(node) *py, BlasFunctions
 				++py;
 			else
 				++px;
-		}			
+		}
 	}
 	return sum;
 }
@@ -483,7 +486,7 @@ double Kernel::k_function(const PREFIX(node) *x, const PREFIX(node) *y,
 				else
 				{
 					if(x->index > y->index)
-					{	
+					{
 						sum += y->value * y->value;
 						++y;
 					}
@@ -520,7 +523,7 @@ double Kernel::k_function(const PREFIX(node) *x, const PREFIX(node) *y,
 #endif
                     }
 		default:
-			return 0;  // Unreachable 
+			return 0;  // Unreachable
 	}
 }
 // An SMO algorithm in Fan et al., JMLR 6(2005), p. 1889--1918
@@ -553,6 +556,7 @@ class Solver {
                 double *upper_bound;
 		double r;	// for Solver_NU
                 bool solve_timed_out;
+		int num_iter;
 	};
 
 	void Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
@@ -597,7 +601,7 @@ class Solver {
 	virtual double calculate_rho();
 	virtual void do_shrinking();
 private:
-	bool be_shrunk(int i, double Gmax1, double Gmax2);	
+	bool be_shrunk(int i, double Gmax1, double Gmax2);
 };
 
 void Solver::swap_index(int i, int j)
@@ -745,11 +749,11 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 			else
 				counter = 1;	// do shrinking next iteration
 		}
-		
+
 		++iter;
 
 		// update alpha[i] and alpha[j], handle bounds carefully
-		
+
 		const Qfloat *Q_i = Q.get_Q(i,active_size);
 		const Qfloat *Q_j = Q.get_Q(j,active_size);
 
@@ -768,7 +772,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 			double diff = alpha[i] - alpha[j];
 			alpha[i] += delta;
 			alpha[j] += delta;
-			
+
 			if(diff > 0)
 			{
 				if(alpha[j] < 0)
@@ -850,7 +854,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 
 		double delta_alpha_i = alpha[i] - old_alpha_i;
 		double delta_alpha_j = alpha[j] - old_alpha_j;
-		
+
 		for(int k=0;k<active_size;k++)
 		{
 			G[k] += Q_i[k]*delta_alpha_i + Q_j[k]*delta_alpha_j;
@@ -919,6 +923,9 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 	for(int i=0;i<l;i++)
 		si->upper_bound[i] = C[i];
 
+	// store number of iterations
+	si->num_iter = iter;
+
 	info("\noptimization finished, #iter = %d\n",iter);
 
 	delete[] p;
@@ -939,7 +946,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 	// j: minimizes the decrease of obj value
 	//    (if quadratic coefficient <= 0, replace it with tau)
 	//    -y_j*grad(f)_j < -y_i*grad(f)_i, j in I_low(\alpha)
-	
+
 	double Gmax = -INF;
 	double Gmax2 = -INF;
 	int Gmax_idx = -1;
@@ -947,7 +954,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 	double obj_diff_min = INF;
 
 	for(int t=0;t<active_size;t++)
-		if(y[t]==+1)	
+		if(y[t]==+1)
 		{
 			if(!is_upper_bound(t))
 				if(-G[t] >= Gmax)
@@ -982,7 +989,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 					Gmax2 = G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff; 
+					double obj_diff;
 					double quad_coef = QD[i]+QD[j]-2.0*y[i]*Q_i[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1006,7 +1013,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 					Gmax2 = -G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff; 
+					double obj_diff;
 					double quad_coef = QD[i]+QD[j]+2.0*y[i]*Q_i[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1044,7 +1051,7 @@ bool Solver::be_shrunk(int i, double Gmax1, double Gmax2)
 	{
 		if(y[i]==+1)
 			return(G[i] > Gmax2);
-		else	
+		else
 			return(G[i] > Gmax1);
 	}
 	else
@@ -1060,27 +1067,27 @@ void Solver::do_shrinking()
 	// find maximal violating pair first
 	for(i=0;i<active_size;i++)
 	{
-		if(y[i]==+1)	
+		if(y[i]==+1)
 		{
-			if(!is_upper_bound(i))	
+			if(!is_upper_bound(i))
 			{
 				if(-G[i] >= Gmax1)
 					Gmax1 = -G[i];
 			}
-			if(!is_lower_bound(i))	
+			if(!is_lower_bound(i))
 			{
 				if(G[i] >= Gmax2)
 					Gmax2 = G[i];
 			}
 		}
-		else	
+		else
 		{
-			if(!is_upper_bound(i))	
+			if(!is_upper_bound(i))
 			{
 				if(-G[i] >= Gmax2)
 					Gmax2 = -G[i];
 			}
-			if(!is_lower_bound(i))	
+			if(!is_lower_bound(i))
 			{
 				if(G[i] >= Gmax1)
 					Gmax1 = G[i];
@@ -1088,7 +1095,7 @@ void Solver::do_shrinking()
 		}
 	}
 
-	if(unshrink == false && Gmax1 + Gmax2 <= eps*10) 
+	if(unshrink == false && Gmax1 + Gmax2 <= eps*10)
 	{
 		unshrink = true;
 		reconstruct_gradient();
@@ -1227,14 +1234,14 @@ int Solver_NU::select_working_set(int &out_i, int &out_j)
 	{
 		if(y[j]==+1)
 		{
-			if (!is_lower_bound(j))	
+			if (!is_lower_bound(j))
 			{
 				double grad_diff=Gmaxp+G[j];
 				if (G[j] >= Gmaxp2)
 					Gmaxp2 = G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff; 
+					double obj_diff;
 					double quad_coef = QD[ip]+QD[j]-2*Q_ip[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1258,7 +1265,7 @@ int Solver_NU::select_working_set(int &out_i, int &out_j)
 					Gmaxn2 = -G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff; 
+					double obj_diff;
 					double quad_coef = QD[in]+QD[j]-2*Q_in[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1293,14 +1300,14 @@ bool Solver_NU::be_shrunk(int i, double Gmax1, double Gmax2, double Gmax3, doubl
 	{
 		if(y[i]==+1)
 			return(-G[i] > Gmax1);
-		else	
+		else
 			return(-G[i] > Gmax4);
 	}
 	else if(is_lower_bound(i))
 	{
 		if(y[i]==+1)
 			return(G[i] > Gmax2);
-		else	
+		else
 			return(G[i] > Gmax3);
 	}
 	else
@@ -1329,14 +1336,14 @@ void Solver_NU::do_shrinking()
 		if(!is_lower_bound(i))
 		{
 			if(y[i]==+1)
-			{	
+			{
 				if(G[i] > Gmax2) Gmax2 = G[i];
 			}
 			else	if(G[i] > Gmax3) Gmax3 = G[i];
 		}
 	}
 
-	if(unshrink == false && max(Gmax1+Gmax2,Gmax3+Gmax4) <= eps*10) 
+	if(unshrink == false && max(Gmax1+Gmax2,Gmax3+Gmax4) <= eps*10)
 	{
 		unshrink = true;
 		reconstruct_gradient();
@@ -1399,12 +1406,12 @@ double Solver_NU::calculate_rho()
 		r1 = sum_free1/nr_free1;
 	else
 		r1 = (ub1+lb1)/2;
-	
+
 	if(nr_free2 > 0)
 		r2 = sum_free2/nr_free2;
 	else
 		r2 = (ub2+lb2)/2;
-	
+
 	si->r = (r1+r2)/2;
 	return (r1-r2)/2;
 }
@@ -1413,7 +1420,7 @@ double Solver_NU::calculate_rho()
 // Q matrices for various formulations
 //
 class SVC_Q: public Kernel
-{ 
+{
 public:
 	SVC_Q(const PREFIX(problem)& prob, const svm_parameter& param, const schar *y_, BlasFunctions *blas_functions)
 	:Kernel(prob.l, prob.x, param, blas_functions)
@@ -1424,7 +1431,7 @@ class SVC_Q: public Kernel
 		for(int i=0;i<prob.l;i++)
 			QD[i] = (this->*kernel_function)(i,i);
 	}
-	
+
 	Qfloat *get_Q(int i, int len) const
 	{
 		Qfloat *data;
@@ -1473,7 +1480,7 @@ class ONE_CLASS_Q: public Kernel
 		for(int i=0;i<prob.l;i++)
 			QD[i] = (this->*kernel_function)(i,i);
 	}
-	
+
 	Qfloat *get_Q(int i, int len) const
 	{
 		Qfloat *data;
@@ -1509,7 +1516,7 @@ class ONE_CLASS_Q: public Kernel
 };
 
 class SVR_Q: public Kernel
-{ 
+{
 public:
 	SVR_Q(const PREFIX(problem)& prob, const svm_parameter& param, BlasFunctions *blas_functions)
 	:Kernel(prob.l, prob.x, param, blas_functions)
@@ -1539,7 +1546,7 @@ class SVR_Q: public Kernel
 		swap(index[i],index[j]);
 		swap(QD[i],QD[j]);
 	}
-	
+
 	Qfloat *get_Q(int i, int len) const
 	{
 		Qfloat *data;
@@ -1655,7 +1662,7 @@ static void solve_nu_svc(
 
 		C[i] = prob->W[i];
 	}
-	
+
 	double nu_l = 0;
 	for(i=0;i<l;i++) nu_l += nu*C[i];
 	double sum_pos = nu_l/2;
@@ -1688,7 +1695,7 @@ static void solve_nu_svc(
 	for(i=0;i<l;i++)
         {
 		alpha[i] *= y[i]/r;
-		si->upper_bound[i] /= r;                
+		si->upper_bound[i] /= r;
         }
 
 	si->rho /= r;
@@ -1836,7 +1843,8 @@ static void solve_nu_svr(
 struct decision_function
 {
 	double *alpha;
-	double rho;	
+	double rho;
+	int num_iter;
 };
 
 static decision_function svm_train_one(
@@ -1848,23 +1856,23 @@ static decision_function svm_train_one(
 	switch(param->svm_type)
 	{
  		case C_SVC:
-			si.upper_bound = Malloc(double,prob->l); 
+			si.upper_bound = Malloc(double,prob->l);
  			solve_c_svc(prob,param,alpha,&si,Cp,Cn,blas_functions);
  			break;
  		case NU_SVC:
-			si.upper_bound = Malloc(double,prob->l); 
+			si.upper_bound = Malloc(double,prob->l);
  			solve_nu_svc(prob,param,alpha,&si,blas_functions);
  			break;
  		case ONE_CLASS:
-			si.upper_bound = Malloc(double,prob->l); 
+			si.upper_bound = Malloc(double,prob->l);
  			solve_one_class(prob,param,alpha,&si,blas_functions);
  			break;
  		case EPSILON_SVR:
-			si.upper_bound = Malloc(double,2*prob->l); 
+			si.upper_bound = Malloc(double,2*prob->l);
  			solve_epsilon_svr(prob,param,alpha,&si,blas_functions);
  			break;
  		case NU_SVR:
-			si.upper_bound = Malloc(double,2*prob->l); 
+			si.upper_bound = Malloc(double,2*prob->l);
  			solve_nu_svr(prob,param,alpha,&si,blas_functions);
  			break;
 	}
@@ -1902,12 +1910,13 @@ static decision_function svm_train_one(
 	decision_function f;
 	f.alpha = alpha;
 	f.rho = si.rho;
+	f.num_iter = si.num_iter;
 	return f;
 }
 
 // Platt's binary SVM Probabilistic Output: an improvement from Lin et al.
 static void sigmoid_train(
-	int l, const double *dec_values, const double *labels, 
+	int l, const double *dec_values, const double *labels,
 	double& A, double& B)
 {
 	double prior1=0, prior0 = 0;
@@ -1916,7 +1925,7 @@ static void sigmoid_train(
 	for (i=0;i<l;i++)
 		if (labels[i] > 0) prior1+=1;
 		else prior0+=1;
-	
+
 	int max_iter=100;	// Maximal number of iterations
 	double min_step=1e-10;	// Minimal step taken in line search
 	double sigma=1e-12;	// For numerically strict PD of Hessian
@@ -1926,8 +1935,8 @@ static void sigmoid_train(
 	double *t=Malloc(double,l);
 	double fApB,p,q,h11,h22,h21,g1,g2,det,dA,dB,gd,stepsize;
 	double newA,newB,newf,d1,d2;
-	int iter; 
-	
+	int iter;
+
 	// Initial Point and Initial Fun Value
 	A=0.0; B=log((prior0+1.0)/(prior1+1.0));
 	double fval = 0.0;
@@ -2037,7 +2046,7 @@ static void multiclass_probability(int k, double **r, double *p)
 	double **Q=Malloc(double *,k);
 	double *Qp=Malloc(double,k);
 	double pQp, eps=0.005/k;
-	
+
 	for (t=0;t<k;t++)
 	{
 		p[t]=1.0/k;  // Valid if k = 1
@@ -2073,7 +2082,7 @@ static void multiclass_probability(int k, double **r, double *p)
 				max_error=error;
 		}
 		if (max_error<eps) break;
-		
+
 		for (t=0;t<k;t++)
 		{
 			double diff=(-Qp[t]+pQp)/Q[t][t];
@@ -2125,7 +2134,7 @@ static void svm_binary_svc_probability(
 #endif
 		subprob.y = Malloc(double,subprob.l);
                 subprob.W = Malloc(double,subprob.l);
-			
+
 		k=0;
 		for(j=0;j<begin;j++)
 		{
@@ -2173,26 +2182,26 @@ static void svm_binary_svc_probability(
 			for(j=begin;j<end;j++)
 			{
 #ifdef _DENSE_REP
-                                PREFIX(predict_values)(submodel,(prob->x+perm[j]),&(dec_values[perm[j]]), blas_functions); 
+                                PREFIX(predict_values)(submodel,(prob->x+perm[j]),&(dec_values[perm[j]]), blas_functions);
 #else
-				PREFIX(predict_values)(submodel,prob->x[perm[j]],&(dec_values[perm[j]]), blas_functions); 
+				PREFIX(predict_values)(submodel,prob->x[perm[j]],&(dec_values[perm[j]]), blas_functions);
 #endif
 				// ensure +1 -1 order; reason not using CV subroutine
 				dec_values[perm[j]] *= submodel->label[0];
-			}		
+			}
 			PREFIX(free_and_destroy_model)(&submodel);
 			PREFIX(destroy_param)(&subparam);
 		}
 		free(subprob.x);
 		free(subprob.y);
                 free(subprob.W);
-	}		
+	}
 	sigmoid_train(prob->l,dec_values,prob->y,probA,probB);
 	free(dec_values);
 	free(perm);
 }
 
-// Return parameter of a Laplace distribution 
+// Return parameter of a Laplace distribution
 static double svm_svr_probability(
 	const PREFIX(problem) *prob, const svm_parameter *param, BlasFunctions *blas_functions)
 {
@@ -2210,15 +2219,15 @@ static double svm_svr_probability(
 	{
 		ymv[i]=prob->y[i]-ymv[i];
 		mae += fabs(ymv[i]);
-	}		
+	}
 	mae /= prob->l;
 	double std=sqrt(2*mae*mae);
 	int count=0;
 	mae=0;
 	for(i=0;i<prob->l;i++)
-		if (fabs(ymv[i]) > 5*std) 
+		if (fabs(ymv[i]) > 5*std)
 			count=count+1;
-		else 
+		else
 			mae+=fabs(ymv[i]);
 	mae /= (prob->l-count);
 	info("Prob. model for test data: target value = predicted value + z,\nz: Laplace distribution e^(-|z|/sigma)/(2sigma),sigma= %g\n",mae);
@@ -2237,7 +2246,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
 	int nr_class = 0;
 	int *label = Malloc(int,max_nr_class);
 	int *count = Malloc(int,max_nr_class);
-	int *data_label = Malloc(int,l);	
+	int *data_label = Malloc(int,l);
 	int i, j, this_label, this_count;
 
 	for(i=0;i<l;i++)
@@ -2265,7 +2274,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
 		}
 	}
 
-        /* 
+        /*
          * Sort labels by straight insertion and apply the same
          * transformation to array count.
          */
@@ -2292,7 +2301,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
                         j ++;
                 }
                 data_label[i] = j;
-        }                
+        }
 
 	int *start = Malloc(int,nr_class);
 	start[0] = 0;
@@ -2319,7 +2328,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
 
 // Remove zero weighed data as libsvm and some liblinear solvers require C > 0.
 //
-static void remove_zero_weight(PREFIX(problem) *newprob, const PREFIX(problem) *prob) 
+static void remove_zero_weight(PREFIX(problem) *newprob, const PREFIX(problem) *prob)
 {
 	int i;
 	int l = 0;
@@ -2376,7 +2385,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		model->probA = NULL; model->probB = NULL;
 		model->sv_coef = Malloc(double *,1);
 
-		if(param->probability && 
+		if(param->probability &&
 		   (param->svm_type == EPSILON_SVR ||
 		    param->svm_type == NU_SVR))
 		{
@@ -2387,6 +2396,8 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
                 NAMESPACE::decision_function f = NAMESPACE::svm_train_one(prob,param,0,0, status,blas_functions);
 		model->rho = Malloc(double,1);
 		model->rho[0] = f.rho;
+		model->num_iter = Malloc(int,1);
+		model->num_iter[0] = f.num_iter;
 
 		int nSV = 0;
 		int i;
@@ -2408,7 +2419,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
                                 model->sv_ind[j] = i;
 				model->sv_coef[0][j] = f.alpha[i];
 				++j;
-			}		
+			}
 
 		free(f.alpha);
 	}
@@ -2423,7 +2434,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		int *perm = Malloc(int,l);
 
 		// group training data of the same class
-                NAMESPACE::svm_group_classes(prob,&nr_class,&label,&start,&count,perm);		
+                NAMESPACE::svm_group_classes(prob,&nr_class,&label,&start,&count,perm);
 #ifdef _DENSE_REP
 		PREFIX(node) *x = Malloc(PREFIX(node),l);
 #else
@@ -2444,7 +2455,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		for(i=0;i<nr_class;i++)
 			weighted_C[i] = param->C;
 		for(i=0;i<param->nr_weight;i++)
-		{	
+		{
 			int j;
 			for(j=0;j<nr_class;j++)
 				if(param->weight_label[i] == label[j])
@@ -2456,7 +2467,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		}
 
 		// train k*(k-1)/2 models
-		
+
 		bool *nonzero = Malloc(bool,l);
 		for(i=0;i<l;i++)
 			nonzero[i] = false;
@@ -2517,14 +2528,18 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		// build output
 
 		model->nr_class = nr_class;
-		
+
 		model->label = Malloc(int,nr_class);
 		for(i=0;i<nr_class;i++)
 			model->label[i] = label[i];
-		
+
 		model->rho = Malloc(double,nr_class*(nr_class-1)/2);
+		model->num_iter = Malloc(int,nr_class*(nr_class-1)/2);
 		for(i=0;i<nr_class*(nr_class-1)/2;i++)
+		{
 			model->rho[i] = f[i].rho;
+			model->num_iter[i] = f[i].num_iter;
+		}
 
 		if(param->probability)
 		{
@@ -2550,7 +2565,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 			int nSV = 0;
 			for(int j=0;j<count[i];j++)
 				if(nonzero[start[i]+j])
-				{	
+				{
 					++nSV;
 					++total_sv;
 				}
@@ -2569,7 +2584,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 #endif
 		p = 0;
 		for(i=0;i<l;i++) {
-			if(nonzero[i]) { 
+			if(nonzero[i]) {
                                 model->SV[p] = x[i];
                                 model->sv_ind[p] = perm[i];
                                 ++p;
@@ -2597,7 +2612,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 				int sj = start[j];
 				int ci = count[i];
 				int cj = count[j];
-				
+
 				int q = nz_start[i];
 				int k;
 				for(k=0;k<ci;k++)
@@ -2609,7 +2624,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 						model->sv_coef[i][q++] = f[p].alpha[ci+k];
 				++p;
 			}
-		
+
 		free(label);
 		free(probA);
 		free(probB);
@@ -2661,7 +2676,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		int *index = Malloc(int,l);
 		for(i=0;i<l;i++)
 			index[i]=perm[i];
-		for (c=0; c<nr_class; c++) 
+		for (c=0; c<nr_class; c++)
 			for(i=0;i<count[c];i++)
 			{
 				int j = i+bounded_rand_int(count[c]-i);
@@ -2690,9 +2705,9 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		fold_start[0]=0;
 		for (i=1;i<=nr_fold;i++)
 			fold_start[i] = fold_start[i-1]+fold_count[i-1];
-		free(start);	
+		free(start);
 		free(label);
-		free(count);	
+		free(count);
 		free(index);
 		free(fold_count);
 	}
@@ -2723,7 +2738,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 #endif
 		subprob.y = Malloc(double,subprob.l);
 		subprob.W = Malloc(double,subprob.l);
-			
+
 		k=0;
 		for(j=0;j<begin;j++)
 		{
@@ -2741,7 +2756,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		}
                 int dummy_status = 0; // IGNORES TIMEOUT ERRORS
 		struct PREFIX(model) *submodel = PREFIX(train)(&subprob,param, &dummy_status, blas_functions);
-		if(param->probability && 
+		if(param->probability &&
 		   (param->svm_type == C_SVC || param->svm_type == NU_SVC))
 		{
 			double *prob_estimates=Malloc(double, PREFIX(get_nr_class)(submodel));
@@ -2751,7 +2766,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 #else
                                 target[perm[j]] = PREFIX(predict_probability)(submodel,prob->x[perm[j]],prob_estimates, blas_functions);
 #endif
-			free(prob_estimates);			
+			free(prob_estimates);
 		}
 		else
 			for(j=begin;j<end;j++)
@@ -2764,9 +2779,9 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		free(subprob.x);
 		free(subprob.y);
                 free(subprob.W);
-	}		
+	}
 	free(fold_start);
-	free(perm);	
+	free(perm);
 }
 
 
@@ -2808,7 +2823,7 @@ double PREFIX(predict_values)(const PREFIX(model) *model, const PREFIX(node) *x,
 	{
 		double *sv_coef = model->sv_coef[0];
 		double sum = 0;
-		
+
 		for(i=0;i<model->l;i++)
 #ifdef _DENSE_REP
                     sum += sv_coef[i] * NAMESPACE::Kernel::k_function(x,model->SV+i,model->param,blas_functions);
@@ -2827,7 +2842,7 @@ double PREFIX(predict_values)(const PREFIX(model) *model, const PREFIX(node) *x,
 	{
 		int nr_class = model->nr_class;
 		int l = model->l;
-		
+
 		double *kvalue = Malloc(double,l);
 		for(i=0;i<l;i++)
 #ifdef _DENSE_REP
@@ -2854,7 +2869,7 @@ double PREFIX(predict_values)(const PREFIX(model) *model, const PREFIX(node) *x,
 				int sj = start[j];
 				int ci = model->nSV[i];
 				int cj = model->nSV[j];
-				
+
 				int k;
 				double *coef1 = model->sv_coef[j-1];
 				double *coef2 = model->sv_coef[i];
@@ -2892,7 +2907,7 @@ double PREFIX(predict)(const PREFIX(model) *model, const PREFIX(node) *x, BlasFu
 	   model->param.svm_type == EPSILON_SVR ||
 	   model->param.svm_type == NU_SVR)
 		dec_values = Malloc(double, 1);
-	else 
+	else
 		dec_values = Malloc(double, nr_class*(nr_class-1)/2);
 	double pred_result = PREFIX(predict_values)(model, x, dec_values, blas_functions);
 	free(dec_values);
@@ -2931,10 +2946,10 @@ double PREFIX(predict_probability)(
 		for(i=0;i<nr_class;i++)
 			free(pairwise_prob[i]);
 		free(dec_values);
-		free(pairwise_prob);	     
+		free(pairwise_prob);
 		return model->label[prob_max_idx];
 	}
-	else 
+	else
 		return PREFIX(predict)(model, x, blas_functions);
 }
 
@@ -2978,6 +2993,9 @@ void PREFIX(free_model_content)(PREFIX(model)* model_ptr)
 
 	free(model_ptr->nSV);
 	model_ptr->nSV = NULL;
+
+	free(model_ptr->num_iter);
+	model_ptr->num_iter = NULL;
 }
 
 void PREFIX(free_and_destroy_model)(PREFIX(model)** model_ptr_ptr)
@@ -3007,9 +3025,9 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 	   svm_type != EPSILON_SVR &&
 	   svm_type != NU_SVR)
 		return "unknown svm type";
-	
+
 	// kernel_type, degree
-	
+
 	int kernel_type = param->kernel_type;
 	if(kernel_type != LINEAR &&
 	   kernel_type != POLY &&
@@ -3062,7 +3080,7 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 
 
 	// check whether nu-svc is feasible
-	
+
 	if(svm_type == NU_SVC)
 	{
 		int l = prob->l;
@@ -3096,7 +3114,7 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 				++nr_class;
 			}
 		}
-	
+
 		for(i=0;i<nr_class;i++)
 		{
 			double n1 = count[i];
@@ -3121,14 +3139,14 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 	   svm_type == ONE_CLASS)
 	{
 		PREFIX(problem) newprob;
-		// filter samples with negative and null weights 
+		// filter samples with negative and null weights
 		remove_zero_weight(&newprob, prob);
 
 		char* msg = NULL;
 		// all samples were removed
 		if(newprob.l == 0)
 			msg =  "Invalid input - all samples have zero or negative weights.";
-		else if(prob->l != newprob.l && 
+		else if(prob->l != newprob.l &&
 		        svm_type == C_SVC)
 		{
 			bool only_one_label = true;
diff --git a/sklearn/svm/src/libsvm/svm.h b/sklearn/svm/src/libsvm/svm.h
index a1634119858f1..44987239e9cf3 100644
--- a/sklearn/svm/src/libsvm/svm.h
+++ b/sklearn/svm/src/libsvm/svm.h
@@ -76,6 +76,7 @@ struct svm_model
 	int l;			/* total #SV */
 	struct svm_node *SV;		/* SVs (SV[l]) */
 	double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
+	int *num_iter;		/* number of iterations run in optimization */
 
 	int *sv_ind;            /* index of support vectors */
 
@@ -101,6 +102,7 @@ struct svm_csr_model
 	int l;			/* total #SV */
 	struct svm_csr_node **SV;		/* SVs (SV[l]) */
 	double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
+	int *num_iter;		/* number of iterations run in optimization */
 
         int *sv_ind;            /* index of support vectors */
 

From 35771a3d3f691c7f245227a81819e8c53983a8a3 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Fri, 22 Oct 2021 10:34:52 -0300
Subject: [PATCH 02/28] Add num_iter to set_model libSVM

---
 sklearn/svm/_base.py                          | 14 ++++--
 sklearn/svm/_libsvm.pxi                       |  2 +-
 sklearn/svm/_libsvm.pyx                       | 24 +++++++--
 sklearn/svm/_libsvm_sparse.pyx                | 17 ++++---
 sklearn/svm/src/libsvm/libsvm_helper.c        |  9 +++-
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c |  9 +++-
 sklearn/svm/tests/test_svm.py                 | 50 +++++++++++++++++--
 7 files changed, 102 insertions(+), 23 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 9adfdc8f821e7..614447e4467a1 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -271,7 +271,9 @@ def fit(self, X, y, sample_weight=None):
         if (
             self._impl in ["c_svc", "nu_svc"] and len(self.classes_) <= 2
         ) or self._impl in ["one_class", "epsilon_svr", "nu_svr"]:
-            self.n_iter_ = self.n_iter_[0]
+            self.n_iter_ = self._num_iter[0]
+        else:
+            self.n_iter_ = self._num_iter
 
         return self
 
@@ -319,7 +321,7 @@ def _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
             self._probA,
             self._probB,
             self.fit_status_,
-            self.n_iter_,
+            self._num_iter,
         ) = libsvm.fit(
             X,
             y,
@@ -360,7 +362,7 @@ def _sparse_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
             self._probA,
             self._probB,
             self.fit_status_,
-            self.n_iter_,
+            self._num_iter,
         ) = libsvm_sparse.libsvm_sparse_train(
             X.shape[1],
             X.data,
@@ -448,6 +450,7 @@ def _dense_predict(self, X):
             self._n_support,
             self._dual_coef_,
             self._intercept_,
+            self._num_iter,
             self._probA,
             self._probB,
             svm_type=svm_type,
@@ -492,6 +495,7 @@ def _sparse_predict(self, X):
             self._n_support,
             self._probA,
             self._probB,
+            self._num_iter,
         )
 
     def _compute_kernel(self, X):
@@ -549,6 +553,7 @@ def _dense_decision_function(self, X):
             self._n_support,
             self._dual_coef_,
             self._intercept_,
+            self._num_iter,
             self._probA,
             self._probB,
             svm_type=LIBSVM_IMPL.index(self._impl),
@@ -592,6 +597,7 @@ def _sparse_decision_function(self, X):
             self._n_support,
             self._probA,
             self._probB,
+            self._num_iter,
         )
 
     def _validate_for_predict(self, X):
@@ -895,6 +901,7 @@ def _dense_predict_proba(self, X):
             self._n_support,
             self._dual_coef_,
             self._intercept_,
+            self._num_iter,
             self._probA,
             self._probB,
             svm_type=svm_type,
@@ -940,6 +947,7 @@ def _sparse_predict_proba(self, X):
             self._n_support,
             self._probA,
             self._probB,
+            self._num_iter,
         )
 
     def _get_coef(self):
diff --git a/sklearn/svm/_libsvm.pxi b/sklearn/svm/_libsvm.pxi
index 46b85f89a0738..9744403eaa85d 100644
--- a/sklearn/svm/_libsvm.pxi
+++ b/sklearn/svm/_libsvm.pxi
@@ -53,7 +53,7 @@ cdef extern from "libsvm_helper.c":
 
     svm_model *set_model (svm_parameter *, int, char *, np.npy_intp *,
                          char *, np.npy_intp *, np.npy_intp *, char *,
-                         char *, char *, char *, char *)
+                         char *, char *, char *, char *, char *)
 
     void copy_sv_coef   (char *, svm_model *)
     void copy_num_iter  (char *, svm_model *)
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index 921fb03373e3b..28fc4edc715d4 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -156,7 +156,7 @@ def fit(
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates, empty array for probability=False.
 
-    n_iter_ : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
+    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
               else (1,)
         Number of iterations run in optimization.
     """
@@ -294,6 +294,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
             np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
             np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+            np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter,
             np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
             np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
             int svm_type=0, kernel='rbf', int degree=3,
@@ -325,6 +326,10 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     intercept : array of shape (n_class*(n_class-1)/2)
         Intercept in decision function.
 
+    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
+              else (1,)
+        Number of iterations run in optimization.
+
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates.
 
@@ -365,7 +370,8 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
                        class_weight_label.data, class_weight.data)
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
-                      sv_coef.data, intercept.data, nSV.data, probA.data, probB.data)
+                      sv_coef.data, intercept.data, nSV.data, probA.data,
+                      probB.data, num_iter.data)
     cdef BlasFunctions blas_functions
     blas_functions.dot = _dot[double]
     #TODO: use check_model
@@ -388,6 +394,7 @@ def predict_proba(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -429,6 +436,10 @@ def predict_proba(
     intercept : array of shape (n_class*(n_class-1)/2,)
         Intercept in decision function.
 
+    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
+              else (1,)
+        Number of iterations run in optimization.
+
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates.
 
@@ -469,7 +480,7 @@ def predict_proba(
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data)
+                      probA.data, probB.data, num_iter.data)
 
     cdef np.npy_intp n_class = get_nr(model)
     cdef BlasFunctions blas_functions
@@ -493,6 +504,7 @@ def decision_function(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -527,6 +539,10 @@ def decision_function(
     intercept : array, shape=[n_class*(n_class-1)/2]
         Intercept in decision function.
 
+    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
+              else (1,)
+        Number of iterations run in optimization.
+
     probA, probB : array, shape=[n_class*(n_class-1)/2]
         Probability estimates.
 
@@ -571,7 +587,7 @@ def decision_function(
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data)
+                      probA.data, probB.data, num_iter.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/_libsvm_sparse.pyx b/sklearn/svm/_libsvm_sparse.pyx
index 3228aa7d7f766..7cf3c6d38f0a5 100644
--- a/sklearn/svm/_libsvm_sparse.pyx
+++ b/sklearn/svm/_libsvm_sparse.pyx
@@ -35,7 +35,7 @@ cdef extern from "libsvm_sparse_helper.c":
                             char *SV_indices, np.npy_intp *SV_intptr_dims,
                             char *SV_intptr,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB)
+                            char *probA, char *probB, char *num_iter)
     svm_parameter *set_parameter (int , int , int , double, double ,
                                   double , double , double , double,
                                   double, int, int, int, char *, char *, int,
@@ -245,7 +245,8 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                             shrinking, int probability,
                             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
                             np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-                            np.ndarray[np.float64_t, ndim=1, mode='c'] probB):
+                            np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
+                            np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter):
     """
     Predict values T given a model.
 
@@ -286,7 +287,7 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data)
+                          nSV.data, probA.data, probB.data, num_iter.data)
     #TODO: use check_model
     dec_values = np.empty(T_indptr.shape[0]-1)
     cdef BlasFunctions blas_functions
@@ -322,7 +323,8 @@ def libsvm_sparse_predict_proba(
     double nu, double p, int shrinking, int probability,
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probB):
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter):
     """
     Predict values T given a model.
     """
@@ -343,7 +345,7 @@ def libsvm_sparse_predict_proba(
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data)
+                          nSV.data, probA.data, probB.data, num_iter.data)
     #TODO: use check_model
     cdef np.npy_intp n_class = get_nr(model)
     cdef int rv
@@ -383,7 +385,8 @@ def libsvm_sparse_decision_function(
     double nu, double p, int shrinking, int probability,
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probB):
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter):
     """
     Predict margin (libsvm name for this is predict_values)
 
@@ -408,7 +411,7 @@ def libsvm_sparse_decision_function(
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data)
+                          nSV.data, probA.data, probB.data, num_iter.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index bfb150235d30b..6280bced31d43 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -109,13 +109,14 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
                             char *support, npy_intp *support_dims,
                             npy_intp *sv_coef_strides,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB)
+                            char *probA, char *probB, char *num_iter)
 {
     struct svm_model *model;
     double *dsv_coef = (double *) sv_coef;
-    int i, m;
+    int i, m, n_models;
 
     m = nr_class * (nr_class-1)/2;
+    n_models = nr_class <= 2 ? 1 : m;
 
     if ((model = malloc(sizeof(struct svm_model))) == NULL)
         goto model_error;
@@ -127,6 +128,8 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
         goto sv_coef_error;
     if ((model->rho = malloc( m * sizeof(double))) == NULL)
         goto rho_error;
+    if ((model->num_iter = malloc(n_models * sizeof(int))) == NULL)
+        goto num_iter_error;
 
     model->nr_class = nr_class;
     model->param = *param;
@@ -181,6 +184,8 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
     model->free_sv = 0;
     return model;
 
+num_iter_error:
+    free(model->num_iter);
 probB_error:
     free(model->probA);
 probA_error:
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index 7234e029b32fd..a9b3066b3ae7e 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -103,13 +103,14 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
                             char *SV_indices, npy_intp *SV_indptr_dims,
                             char *SV_intptr,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB)
+                            char *probA, char *probB, char *num_iter)
 {
     struct svm_csr_model *model;
     double *dsv_coef = (double *) sv_coef;
-    int i, m;
+    int i, m, n_models;
 
     m = nr_class * (nr_class-1)/2;
+    n_models = nr_class <= 2 ? 1 : m;
 
     if ((model = malloc(sizeof(struct svm_csr_model))) == NULL)
         goto model_error;
@@ -121,6 +122,8 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
         goto sv_coef_error;
     if ((model->rho = malloc( m * sizeof(double))) == NULL)
         goto rho_error;
+    if ((model->num_iter = malloc(n_models * sizeof(int))) == NULL)
+        goto num_iter_error;
 
     /* in the case of precomputed kernels we do not use
        dense_to_precomputed because we don't want the leading 0. As
@@ -180,6 +183,8 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
     model->free_sv = 0;
     return model;
 
+num_iter_error:
+    free(model->num_iter);
 probB_error:
     free(model->probA);
 probA_error:
diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index eb15aca0096b8..cf777cc4a2b1d 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -66,12 +66,54 @@ def test_libsvm_iris():
     assert_array_equal(clf.classes_, np.sort(clf.classes_))
 
     # check also the low-level API
-    model = _libsvm.fit(iris.data, iris.target.astype(np.float64))
-    pred = _libsvm.predict(iris.data, *model)
+    (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_probA,
+        libsvm_probB,
+        libsvm_fit_status,
+        libsvm_num_iter,
+    ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
+
+    lib_svm_model = (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_num_iter,
+        libsvm_probA,
+        libsvm_probB,
+    )
+    pred = _libsvm.predict(iris.data, *lib_svm_model)
     assert np.mean(pred == iris.target) > 0.95
 
-    model = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
-    pred = _libsvm.predict(iris.data, *model, kernel="linear")
+    (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_probA,
+        libsvm_probB,
+        libsvm_fit_status,
+        libsvm_num_iter,
+    ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
+
+    lib_svm_model = (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_num_iter,
+        libsvm_probA,
+        libsvm_probB,
+    )
+    pred = _libsvm.predict(iris.data, *lib_svm_model, kernel="linear")
     assert np.mean(pred == iris.target) > 0.95
 
     pred = _libsvm.cross_validation(

From ffb657c85fa2852feed4414d2cd36e8e46ade7a1 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Fri, 22 Oct 2021 15:26:24 -0300
Subject: [PATCH 03/28] Add changelog

---
 doc/whats_new/v1.1.rst | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/doc/whats_new/v1.1.rst b/doc/whats_new/v1.1.rst
index 2bd0246efad15..e6bf89e63632e 100644
--- a/doc/whats_new/v1.1.rst
+++ b/doc/whats_new/v1.1.rst
@@ -105,6 +105,12 @@ Changelog
   Setting a transformer to "passthrough" will pass the features unchanged.
   :pr:`20860` by :user:`Shubhraneel Pal <shubhraneel>`.
 
+:mod:`sklearn.svm`
+...................
+
+- |Enhancement| Exposed the number of iterations attribute in
+  :class:`svm.BaseLibSVM`. :pr:`21408` by :user:`Juan Martín Loyola <jmloyola>`.
+
 :mod:`sklearn.utils`
 ....................
 

From 465ade09ef0f2c66b340775aef2a8b772d286993 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Fri, 22 Oct 2021 15:28:26 -0300
Subject: [PATCH 04/28] Test n_iter_ for BaseLibSVM estimators

---
 sklearn/utils/estimator_checks.py | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/sklearn/utils/estimator_checks.py b/sklearn/utils/estimator_checks.py
index ccc6ff23ed8fc..f5dc94fc42fe4 100644
--- a/sklearn/utils/estimator_checks.py
+++ b/sklearn/utils/estimator_checks.py
@@ -3217,11 +3217,7 @@ def check_non_transformer_estimators_n_iter(name, estimator_orig):
     # labeled, hence n_iter_ = 0 is valid.
     not_run_check_n_iter = [
         "Ridge",
-        "SVR",
-        "NuSVR",
-        "NuSVC",
         "RidgeClassifier",
-        "SVC",
         "RandomizedLasso",
         "LogisticRegressionCV",
         "LinearSVC",
@@ -3248,7 +3244,11 @@ def check_non_transformer_estimators_n_iter(name, estimator_orig):
 
         estimator.fit(X, y_)
 
-        assert estimator.n_iter_ >= 1
+        # These return a n_iter per model optimized
+        if name in ["SVC", "NuSVC"] and len(estimator.classes_) > 2:
+            assert np.all(estimator.n_iter_ >= 1)
+        else:
+            assert estimator.n_iter_ >= 1
 
 
 @ignore_warnings(category=FutureWarning)
@@ -3275,7 +3275,9 @@ def check_transformer_n_iter(name, estimator_orig):
         estimator.fit(X, y_)
 
         # These return a n_iter per component.
-        if name in CROSS_DECOMPOSITION:
+        if (name in CROSS_DECOMPOSITION) or (
+            name in ["SVC", "NuSVC"] and len(estimator.classes_) > 2
+        ):
             for iter_ in estimator.n_iter_:
                 assert iter_ >= 1
         else:

From 347923136123fe03e3da4a5743f58ad6cf48df91 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Sun, 24 Oct 2021 23:04:10 -0300
Subject: [PATCH 05/28] Fix failing test

---
 sklearn/utils/estimator_checks.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/sklearn/utils/estimator_checks.py b/sklearn/utils/estimator_checks.py
index f5dc94fc42fe4..92e5c1622270a 100644
--- a/sklearn/utils/estimator_checks.py
+++ b/sklearn/utils/estimator_checks.py
@@ -3242,6 +3242,9 @@ def check_non_transformer_estimators_n_iter(name, estimator_orig):
 
         set_random_state(estimator, 0)
 
+        if name in ["SVC", "NuSVC"] and estimator.kernel == "precomputed":
+            X = np.dot(X, X.T)
+
         estimator.fit(X, y_)
 
         # These return a n_iter per model optimized

From ccddcb4b6f02ef30fbeb5557d882a58bb3fe7a09 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 26 Oct 2021 16:06:49 -0300
Subject: [PATCH 06/28] Apply suggestions from code review

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 sklearn/svm/src/libsvm/LIBSVM_CHANGES | 2 +-
 sklearn/svm/src/libsvm/svm.cpp        | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/sklearn/svm/src/libsvm/LIBSVM_CHANGES b/sklearn/svm/src/libsvm/LIBSVM_CHANGES
index 7d594c826e5f5..663550b8ddd6f 100644
--- a/sklearn/svm/src/libsvm/LIBSVM_CHANGES
+++ b/sklearn/svm/src/libsvm/LIBSVM_CHANGES
@@ -7,5 +7,5 @@ This is here mainly as checklist for incorporation of new versions of libsvm.
   * Improved random number generator (fix on windows, enhancement on other
     platforms). See <https://github.com/scikit-learn/scikit-learn/pull/13511#issuecomment-481729756>
   * invoke scipy blas api for svm kernel function to improve performance with speedup rate of 1.5X to 2X for dense data only. See <https://github.com/scikit-learn/scikit-learn/pull/16530>
-  * Expose the number of iterations run in optimization.
+  * Expose the number of iterations run in optimization. See <https://github.com/scikit-learn/scikit-learn/pull/21408>
 The changes made with respect to upstream are detailed in the heading of svm.cpp
diff --git a/sklearn/svm/src/libsvm/svm.cpp b/sklearn/svm/src/libsvm/svm.cpp
index 84e9ca09c44a2..5d0bb5baef93c 100644
--- a/sklearn/svm/src/libsvm/svm.cpp
+++ b/sklearn/svm/src/libsvm/svm.cpp
@@ -58,6 +58,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
    Modified 2021:
 
    - Exposed number of iterations run in optimization, Juan Martín Loyola.
+     See <https://github.com/scikit-learn/scikit-learn/pull/21408/>
  */
 
 #include <math.h>

From 23e16feb704871cf5479c35d66a002ef3f6b4ab9 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 26 Oct 2021 16:16:19 -0300
Subject: [PATCH 07/28] Apply reviewers suggestions

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 doc/whats_new/v1.1.rst | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/doc/whats_new/v1.1.rst b/doc/whats_new/v1.1.rst
index 850298222a3bc..15c9af876b3e6 100644
--- a/doc/whats_new/v1.1.rst
+++ b/doc/whats_new/v1.1.rst
@@ -91,8 +91,10 @@ Changelog
 :mod:`sklearn.svm`
 ...................
 
-- |Enhancement| Exposed the number of iterations attribute in
-  :class:`svm.BaseLibSVM`. :pr:`21408` by :user:`Juan Martín Loyola <jmloyola>`.
+- |Enhancement| :class:`svm.OneClassSVM`, :class:`svm.NuSVC`,
+  :class:`svm.NuSVR`, :class:`svm.SVC` and :class:`svm.SVR` now expose
+  `n_iter_`, the number of iterations used by the libsvm optimization routine.
+  :pr:`21408` by :user:`Juan Martín Loyola <jmloyola>`.
 
 :mod:`sklearn.utils`
 ....................

From b2a15842d38c30f0c5526483617126dc3c5366b1 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 26 Oct 2021 17:13:28 -0300
Subject: [PATCH 08/28] Apply reviewers suggestions

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 sklearn/svm/_classes.py        | 30 +++++++++++++++---------------
 sklearn/svm/_libsvm.pyx        | 28 ++++++++++------------------
 sklearn/svm/_libsvm_sparse.pyx |  8 ++------
 sklearn/svm/src/libsvm/svm.h   |  4 ++--
 4 files changed, 29 insertions(+), 41 deletions(-)

diff --git a/sklearn/svm/_classes.py b/sklearn/svm/_classes.py
index 0172fe6c6e983..4f4bce25f88c3 100644
--- a/sklearn/svm/_classes.py
+++ b/sklearn/svm/_classes.py
@@ -670,12 +670,12 @@ class SVC(BaseSVC):
 
         .. versionadded:: 1.0
 
-    n_iter_ : int or ndarray of shape (n_class*(n_class-1)/2,)
-        Number of iterations run in optimization. If `classes_ <= 2`, only one
-        model is optimized, thus an integer is returned. Otherwise, multiple
-        models are optimized separately, thus having multiple number of
-        iterations. In this case a numpy array is returned with the number of
-        iterations for each model.
+    n_iter_ : int or ndarray of shape (n_class * (n_class - 1) // 2,)
+        Number of iterations run by the optimization routine to fit the model.
+        If `classes_ <= 2`, only one model is optimized, thus an integer is
+        returned. Otherwise, multiple models are optimized separately, thus
+        having multiple number of iterations. In this case a numpy array is
+        returned with the number of iterations for each model.
 
         .. versionadded:: 1.1
 
@@ -934,12 +934,12 @@ class NuSVC(BaseSVC):
 
         .. versionadded:: 1.0
 
-    n_iter_ : int or ndarray of shape (n_class*(n_class-1)/2,)
-        Number of iterations run in optimization. If `classes_ <= 2`, only one
-        model is optimized, thus an integer is returned. Otherwise, multiple
-        models are optimized separately, thus having multiple number of
-        iterations. In this case a numpy array is returned with the number of
-        iterations for each model.
+    n_iter_ : int or ndarray of shape (n_class * (n_class - 1) // 2,)
+        Number of iterations run by the optimization routine to fit the model.
+        If `classes_ <= 2`, only one model is optimized, thus an integer is
+        returned. Otherwise, multiple models are optimized separately, thus
+        having multiple number of iterations. In this case a numpy array is
+        returned with the number of iterations for each model.
 
         .. versionadded:: 1.1
 
@@ -1159,7 +1159,7 @@ class SVR(RegressorMixin, BaseLibSVM):
         .. versionadded:: 1.0
 
     n_iter_ : int
-        Number of iterations run in optimization.
+        Number of iterations run by the optimization routine to fit the model.
 
         .. versionadded:: 1.1
 
@@ -1352,7 +1352,7 @@ class NuSVR(RegressorMixin, BaseLibSVM):
         .. versionadded:: 1.0
 
     n_iter_ : int
-        Number of iterations run in optimization.
+        Number of iterations run by the optimization routine to fit the model.
 
         .. versionadded:: 1.1
 
@@ -1541,7 +1541,7 @@ class OneClassSVM(OutlierMixin, BaseLibSVM):
         .. versionadded:: 1.0
 
     n_iter_ : int
-        Number of iterations run in optimization.
+        Number of iterations run by the optimization routine to fit the model.
 
         .. versionadded:: 1.1
 
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index 28fc4edc715d4..a123e364de72a 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -156,9 +156,8 @@ def fit(
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates, empty array for probability=False.
 
-    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
-              else (1,)
-        Number of iterations run in optimization.
+    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+        Number of iterations run by the optimization routine to fit the model.
     """
 
     cdef svm_parameter param
@@ -207,12 +206,8 @@ def fit(
     n_class = get_nr(model)
 
     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter
-    if n_class > 2:
-        num_iter = np.empty (int((n_class*(n_class-1))/2), dtype=np.int32)
-        copy_num_iter (num_iter.data, model)
-    else:
-        num_iter = np.empty (1, dtype=np.int32)
-        copy_num_iter (num_iter.data, model)
+    num_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
+    copy_num_iter (num_iter.data, model)
 
     cdef np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef
     sv_coef = np.empty((n_class-1, SV_len), dtype=np.float64)
@@ -326,9 +321,8 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     intercept : array of shape (n_class*(n_class-1)/2)
         Intercept in decision function.
 
-    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
-              else (1,)
-        Number of iterations run in optimization.
+    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+        Number of iterations run by the optimization routine to fit the model.
 
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates.
@@ -436,9 +430,8 @@ def predict_proba(
     intercept : array of shape (n_class*(n_class-1)/2,)
         Intercept in decision function.
 
-    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
-              else (1,)
-        Number of iterations run in optimization.
+    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+        Number of iterations run by the optimization routine to fit the model.
 
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates.
@@ -539,9 +532,8 @@ def decision_function(
     intercept : array, shape=[n_class*(n_class-1)/2]
         Intercept in decision function.
 
-    num_iter : ndarray of shape (n_class*(n_class-1)/2,) if n_class > 2 \
-              else (1,)
-        Number of iterations run in optimization.
+    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+        Number of iterations run by the optimization routine to fit the model.
 
     probA, probB : array, shape=[n_class*(n_class-1)/2]
         Probability estimates.
diff --git a/sklearn/svm/_libsvm_sparse.pyx b/sklearn/svm/_libsvm_sparse.pyx
index 7cf3c6d38f0a5..21207fc81c66f 100644
--- a/sklearn/svm/_libsvm_sparse.pyx
+++ b/sklearn/svm/_libsvm_sparse.pyx
@@ -161,12 +161,8 @@ def libsvm_sparse_train ( int n_features,
     cdef np.npy_intp n_class = get_nr(model)
 
     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter
-    if n_class > 2:
-        num_iter = np.empty (int((n_class*(n_class-1))/2), dtype=np.int32)
-        copy_num_iter (num_iter.data, model)
-    else:
-        num_iter = np.empty (1, dtype=np.int32)
-        copy_num_iter (num_iter.data, model)
+    num_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
+    copy_num_iter (num_iter.data, model)
 
     # copy model.sv_coef
     # we create a new array instead of resizing, otherwise
diff --git a/sklearn/svm/src/libsvm/svm.h b/sklearn/svm/src/libsvm/svm.h
index 44987239e9cf3..18c0671fe239d 100644
--- a/sklearn/svm/src/libsvm/svm.h
+++ b/sklearn/svm/src/libsvm/svm.h
@@ -76,7 +76,7 @@ struct svm_model
 	int l;			/* total #SV */
 	struct svm_node *SV;		/* SVs (SV[l]) */
 	double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
-	int *num_iter;		/* number of iterations run in optimization */
+	int *num_iter;		/* number of iterations run by the optimization routine to fit the model */
 
 	int *sv_ind;            /* index of support vectors */
 
@@ -102,7 +102,7 @@ struct svm_csr_model
 	int l;			/* total #SV */
 	struct svm_csr_node **SV;		/* SVs (SV[l]) */
 	double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
-	int *num_iter;		/* number of iterations run in optimization */
+	int *num_iter;		/* number of iterations run by the optimization routine to fit the model */
 
         int *sv_ind;            /* index of support vectors */
 

From a6b6151bb3415126d18c9be8dd1091652a366b8a Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 26 Oct 2021 21:34:56 -0300
Subject: [PATCH 09/28] Apply reviewers suggestions

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
---
 sklearn/svm/_libsvm.pxi                       |  2 +-
 sklearn/svm/_libsvm.pyx                       | 28 +++++++++----------
 sklearn/svm/_libsvm_sparse.pyx                | 24 ++++++++--------
 sklearn/svm/src/libsvm/libsvm_helper.c        | 16 +++++------
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c | 16 +++++------
 sklearn/svm/src/libsvm/svm.cpp                | 20 ++++++-------
 sklearn/svm/src/libsvm/svm.h                  |  4 +--
 sklearn/svm/tests/test_svm.py                 |  8 +++---
 8 files changed, 59 insertions(+), 59 deletions(-)

diff --git a/sklearn/svm/_libsvm.pxi b/sklearn/svm/_libsvm.pxi
index 9744403eaa85d..39cc40a0bc9ae 100644
--- a/sklearn/svm/_libsvm.pxi
+++ b/sklearn/svm/_libsvm.pxi
@@ -56,7 +56,7 @@ cdef extern from "libsvm_helper.c":
                          char *, char *, char *, char *, char *)
 
     void copy_sv_coef   (char *, svm_model *)
-    void copy_num_iter  (char *, svm_model *)
+    void copy_n_iter  (char *, svm_model *)
     void copy_intercept (char *, svm_model *, np.npy_intp *)
     void copy_SV        (char *, svm_model *, np.npy_intp *)
     int copy_support (char *data, svm_model *model)
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index a123e364de72a..aa55e6ba3a3d2 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -156,7 +156,7 @@ def fit(
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates, empty array for probability=False.
 
-    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
         Number of iterations run by the optimization routine to fit the model.
     """
 
@@ -205,9 +205,9 @@ def fit(
     SV_len  = get_l(model)
     n_class = get_nr(model)
 
-    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter
-    num_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
-    copy_num_iter (num_iter.data, model)
+    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter
+    n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
+    copy_n_iter(n_iter.data, model)
 
     cdef np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef
     sv_coef = np.empty((n_class-1, SV_len), dtype=np.float64)
@@ -258,7 +258,7 @@ def fit(
     free(problem.x)
 
     return (support, support_vectors, n_class_SV, sv_coef, intercept,
-           probA, probB, fit_status, num_iter)
+           probA, probB, fit_status, n_iter)
 
 
 cdef void set_predict_params(
@@ -289,7 +289,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
             np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
             np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-            np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter,
+            np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter,
             np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
             np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
             int svm_type=0, kernel='rbf', int degree=3,
@@ -321,7 +321,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     intercept : array of shape (n_class*(n_class-1)/2)
         Intercept in decision function.
 
-    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
         Number of iterations run by the optimization routine to fit the model.
 
     probA, probB : array of shape (n_class*(n_class-1)/2,)
@@ -365,7 +365,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data, probA.data,
-                      probB.data, num_iter.data)
+                      probB.data, n_iter.data)
     cdef BlasFunctions blas_functions
     blas_functions.dot = _dot[double]
     #TODO: use check_model
@@ -388,7 +388,7 @@ def predict_proba(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -430,7 +430,7 @@ def predict_proba(
     intercept : array of shape (n_class*(n_class-1)/2,)
         Intercept in decision function.
 
-    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
         Number of iterations run by the optimization routine to fit the model.
 
     probA, probB : array of shape (n_class*(n_class-1)/2,)
@@ -473,7 +473,7 @@ def predict_proba(
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data, num_iter.data)
+                      probA.data, probB.data, n_iter.data)
 
     cdef np.npy_intp n_class = get_nr(model)
     cdef BlasFunctions blas_functions
@@ -497,7 +497,7 @@ def decision_function(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -532,7 +532,7 @@ def decision_function(
     intercept : array, shape=[n_class*(n_class-1)/2]
         Intercept in decision function.
 
-    num_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
+    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
         Number of iterations run by the optimization routine to fit the model.
 
     probA, probB : array, shape=[n_class*(n_class-1)/2]
@@ -579,7 +579,7 @@ def decision_function(
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data, num_iter.data)
+                      probA.data, probB.data, n_iter.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/_libsvm_sparse.pyx b/sklearn/svm/_libsvm_sparse.pyx
index 21207fc81c66f..f21e9daa6d962 100644
--- a/sklearn/svm/_libsvm_sparse.pyx
+++ b/sklearn/svm/_libsvm_sparse.pyx
@@ -35,13 +35,13 @@ cdef extern from "libsvm_sparse_helper.c":
                             char *SV_indices, np.npy_intp *SV_intptr_dims,
                             char *SV_intptr,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB, char *num_iter)
+                            char *probA, char *probB, char *n_iter)
     svm_parameter *set_parameter (int , int , int , double, double ,
                                   double , double , double , double,
                                   double, int, int, int, char *, char *, int,
                                   int)
     void copy_sv_coef   (char *, svm_csr_model *)
-    void copy_num_iter  (char *, svm_csr_model *)
+    void copy_n_iter  (char *, svm_csr_model *)
     void copy_support   (char *, svm_csr_model *)
     void copy_intercept (char *, svm_csr_model *, np.npy_intp *)
     int copy_predict (char *, svm_csr_model *, np.npy_intp *, char *, BlasFunctions *)
@@ -160,9 +160,9 @@ def libsvm_sparse_train ( int n_features,
     cdef np.npy_intp SV_len = get_l(model)
     cdef np.npy_intp n_class = get_nr(model)
 
-    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter
-    num_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
-    copy_num_iter (num_iter.data, model)
+    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter
+    n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
+    copy_n_iter(n_iter.data, model)
 
     # copy model.sv_coef
     # we create a new array instead of resizing, otherwise
@@ -222,7 +222,7 @@ def libsvm_sparse_train ( int n_features,
     free_param(param)
 
     return (support, support_vectors_, sv_coef_data, intercept, n_class_SV,
-            probA, probB, fit_status, num_iter)
+            probA, probB, fit_status, n_iter)
 
 
 def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
@@ -242,7 +242,7 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
                             np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
                             np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-                            np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter):
+                            np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter):
     """
     Predict values T given a model.
 
@@ -283,7 +283,7 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data, num_iter.data)
+                          nSV.data, probA.data, probB.data, n_iter.data)
     #TODO: use check_model
     dec_values = np.empty(T_indptr.shape[0]-1)
     cdef BlasFunctions blas_functions
@@ -320,7 +320,7 @@ def libsvm_sparse_predict_proba(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter):
+    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter):
     """
     Predict values T given a model.
     """
@@ -341,7 +341,7 @@ def libsvm_sparse_predict_proba(
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data, num_iter.data)
+                          nSV.data, probA.data, probB.data, n_iter.data)
     #TODO: use check_model
     cdef np.npy_intp n_class = get_nr(model)
     cdef int rv
@@ -382,7 +382,7 @@ def libsvm_sparse_decision_function(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] num_iter):
+    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter):
     """
     Predict margin (libsvm name for this is predict_values)
 
@@ -407,7 +407,7 @@ def libsvm_sparse_decision_function(
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data, num_iter.data)
+                          nSV.data, probA.data, probB.data, n_iter.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index 6280bced31d43..b91cb8870d2c9 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -109,7 +109,7 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
                             char *support, npy_intp *support_dims,
                             npy_intp *sv_coef_strides,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB, char *num_iter)
+                            char *probA, char *probB, char *n_iter)
 {
     struct svm_model *model;
     double *dsv_coef = (double *) sv_coef;
@@ -128,8 +128,8 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
         goto sv_coef_error;
     if ((model->rho = malloc( m * sizeof(double))) == NULL)
         goto rho_error;
-    if ((model->num_iter = malloc(n_models * sizeof(int))) == NULL)
-        goto num_iter_error;
+    if ((model->n_iter = malloc(n_models * sizeof(int))) == NULL)
+        goto n_iter_error;
 
     model->nr_class = nr_class;
     model->param = *param;
@@ -184,8 +184,8 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
     model->free_sv = 0;
     return model;
 
-num_iter_error:
-    free(model->num_iter);
+n_iter_error:
+    free(model->n_iter);
 probB_error:
     free(model->probA);
 probA_error:
@@ -226,11 +226,11 @@ npy_intp get_nr(struct svm_model *model)
 /*
  * Get the number of iterations run in optimization
  */
-void copy_num_iter(char *data, struct svm_model *model)
+void copy_n_iter(char *data, struct svm_model *model)
 {
     int n_models;
     n_models = model->nr_class <= 2 ? 1 : model->nr_class * (model->nr_class-1)/2;
-    memcpy(data, model->num_iter, n_models * sizeof(int));
+    memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 
 /*
@@ -386,7 +386,7 @@ int free_model(struct svm_model *model)
     free(model->label);
     free(model->probA);
     free(model->probB);
-    free(model->num_iter);
+    free(model->n_iter);
     free(model->nSV);
     free(model);
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index a9b3066b3ae7e..2779bfbfeed73 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -103,7 +103,7 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
                             char *SV_indices, npy_intp *SV_indptr_dims,
                             char *SV_intptr,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB, char *num_iter)
+                            char *probA, char *probB, char *n_iter)
 {
     struct svm_csr_model *model;
     double *dsv_coef = (double *) sv_coef;
@@ -122,8 +122,8 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
         goto sv_coef_error;
     if ((model->rho = malloc( m * sizeof(double))) == NULL)
         goto rho_error;
-    if ((model->num_iter = malloc(n_models * sizeof(int))) == NULL)
-        goto num_iter_error;
+    if ((model->n_iter = malloc(n_models * sizeof(int))) == NULL)
+        goto n_iter_error;
 
     /* in the case of precomputed kernels we do not use
        dense_to_precomputed because we don't want the leading 0. As
@@ -183,8 +183,8 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
     model->free_sv = 0;
     return model;
 
-num_iter_error:
-    free(model->num_iter);
+n_iter_error:
+    free(model->n_iter);
 probB_error:
     free(model->probA);
 probA_error:
@@ -356,11 +356,11 @@ void copy_sv_coef(char *data, struct svm_csr_model *model)
 /*
  * Get the number of iterations run in optimization
  */
-void copy_num_iter(char *data, struct svm_csr_model *model)
+void copy_n_iter(char *data, struct svm_csr_model *model)
 {
     int n_models;
     n_models = model->nr_class <= 2 ? 1 : model->nr_class * (model->nr_class-1)/2;
-    memcpy(data, model->num_iter, n_models * sizeof(int));
+    memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 
 /*
@@ -424,7 +424,7 @@ int free_model(struct svm_csr_model *model)
     free(model->label);
     free(model->probA);
     free(model->probB);
-    free(model->num_iter);
+    free(model->n_iter);
     free(model->nSV);
     free(model);
 
diff --git a/sklearn/svm/src/libsvm/svm.cpp b/sklearn/svm/src/libsvm/svm.cpp
index 5d0bb5baef93c..e841296cbbcc2 100644
--- a/sklearn/svm/src/libsvm/svm.cpp
+++ b/sklearn/svm/src/libsvm/svm.cpp
@@ -557,7 +557,7 @@ class Solver {
                 double *upper_bound;
 		double r;	// for Solver_NU
                 bool solve_timed_out;
-		int num_iter;
+		int n_iter;
 	};
 
 	void Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
@@ -925,7 +925,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 		si->upper_bound[i] = C[i];
 
 	// store number of iterations
-	si->num_iter = iter;
+	si->n_iter = iter;
 
 	info("\noptimization finished, #iter = %d\n",iter);
 
@@ -1845,7 +1845,7 @@ struct decision_function
 {
 	double *alpha;
 	double rho;
-	int num_iter;
+	int n_iter;
 };
 
 static decision_function svm_train_one(
@@ -1911,7 +1911,7 @@ static decision_function svm_train_one(
 	decision_function f;
 	f.alpha = alpha;
 	f.rho = si.rho;
-	f.num_iter = si.num_iter;
+	f.n_iter = si.n_iter;
 	return f;
 }
 
@@ -2397,8 +2397,8 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
                 NAMESPACE::decision_function f = NAMESPACE::svm_train_one(prob,param,0,0, status,blas_functions);
 		model->rho = Malloc(double,1);
 		model->rho[0] = f.rho;
-		model->num_iter = Malloc(int,1);
-		model->num_iter[0] = f.num_iter;
+		model->n_iter = Malloc(int,1);
+		model->n_iter[0] = f.n_iter;
 
 		int nSV = 0;
 		int i;
@@ -2535,11 +2535,11 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 			model->label[i] = label[i];
 
 		model->rho = Malloc(double,nr_class*(nr_class-1)/2);
-		model->num_iter = Malloc(int,nr_class*(nr_class-1)/2);
+		model->n_iter = Malloc(int,nr_class*(nr_class-1)/2);
 		for(i=0;i<nr_class*(nr_class-1)/2;i++)
 		{
 			model->rho[i] = f[i].rho;
-			model->num_iter[i] = f[i].num_iter;
+			model->n_iter[i] = f[i].n_iter;
 		}
 
 		if(param->probability)
@@ -2995,8 +2995,8 @@ void PREFIX(free_model_content)(PREFIX(model)* model_ptr)
 	free(model_ptr->nSV);
 	model_ptr->nSV = NULL;
 
-	free(model_ptr->num_iter);
-	model_ptr->num_iter = NULL;
+	free(model_ptr->n_iter);
+	model_ptr->n_iter = NULL;
 }
 
 void PREFIX(free_and_destroy_model)(PREFIX(model)** model_ptr_ptr)
diff --git a/sklearn/svm/src/libsvm/svm.h b/sklearn/svm/src/libsvm/svm.h
index 18c0671fe239d..518872c67bc5c 100644
--- a/sklearn/svm/src/libsvm/svm.h
+++ b/sklearn/svm/src/libsvm/svm.h
@@ -76,7 +76,7 @@ struct svm_model
 	int l;			/* total #SV */
 	struct svm_node *SV;		/* SVs (SV[l]) */
 	double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
-	int *num_iter;		/* number of iterations run by the optimization routine to fit the model */
+	int *n_iter;		/* number of iterations run by the optimization routine to fit the model */
 
 	int *sv_ind;            /* index of support vectors */
 
@@ -102,7 +102,7 @@ struct svm_csr_model
 	int l;			/* total #SV */
 	struct svm_csr_node **SV;		/* SVs (SV[l]) */
 	double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
-	int *num_iter;		/* number of iterations run by the optimization routine to fit the model */
+	int *n_iter;		/* number of iterations run by the optimization routine to fit the model */
 
         int *sv_ind;            /* index of support vectors */
 
diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index cf777cc4a2b1d..17067183c9197 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -75,7 +75,7 @@ def test_libsvm_iris():
         libsvm_probA,
         libsvm_probB,
         libsvm_fit_status,
-        libsvm_num_iter,
+        libsvm_n_iter,
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
 
     lib_svm_model = (
@@ -84,7 +84,7 @@ def test_libsvm_iris():
         libsvm_n_class_SV,
         libsvm_sv_coef,
         libsvm_intercept,
-        libsvm_num_iter,
+        libsvm_n_iter,
         libsvm_probA,
         libsvm_probB,
     )
@@ -100,7 +100,7 @@ def test_libsvm_iris():
         libsvm_probA,
         libsvm_probB,
         libsvm_fit_status,
-        libsvm_num_iter,
+        libsvm_n_iter,
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
 
     lib_svm_model = (
@@ -109,7 +109,7 @@ def test_libsvm_iris():
         libsvm_n_class_SV,
         libsvm_sv_coef,
         libsvm_intercept,
-        libsvm_num_iter,
+        libsvm_n_iter,
         libsvm_probA,
         libsvm_probB,
     )

From bbc4432d1a72cf3f45ec7ca4ad9e0b65fb534a9c Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Wed, 27 Oct 2021 11:44:42 -0300
Subject: [PATCH 10/28] Revert remotion of trailing white spaces

---
 sklearn/svm/src/libsvm/svm.cpp | 192 ++++++++++++++++-----------------
 1 file changed, 96 insertions(+), 96 deletions(-)

diff --git a/sklearn/svm/src/libsvm/svm.cpp b/sklearn/svm/src/libsvm/svm.cpp
index e841296cbbcc2..de07fecdba2ac 100644
--- a/sklearn/svm/src/libsvm/svm.cpp
+++ b/sklearn/svm/src/libsvm/svm.cpp
@@ -31,7 +31,7 @@ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-/*
+/* 
    Modified 2010:
 
    - Support for dense data by Ming-Fang Weng
@@ -129,7 +129,7 @@ static void info(const char *fmt,...)
 and dense versions of this library */
 #ifdef _DENSE_REP
   #ifdef PREFIX
-    #undef PREFIX
+    #undef PREFIX  
   #endif
   #ifdef NAMESPACE
     #undef NAMESPACE
@@ -140,7 +140,7 @@ and dense versions of this library */
 #else
   /* sparse representation */
   #ifdef PREFIX
-    #undef PREFIX
+    #undef PREFIX  
   #endif
   #ifdef NAMESPACE
     #undef NAMESPACE
@@ -167,7 +167,7 @@ class Cache
 	// return some position p where [p,len) need to be filled
 	// (p >= len if nothing needs to be filled)
 	int get_data(const int index, Qfloat **data, int len);
-	void swap_index(int i, int j);
+	void swap_index(int i, int j);	
 private:
 	int l;
 	long int size;
@@ -443,7 +443,7 @@ double Kernel::dot(const PREFIX(node) *px, const PREFIX(node) *py, BlasFunctions
 				++py;
 			else
 				++px;
-		}
+		}			
 	}
 	return sum;
 }
@@ -487,7 +487,7 @@ double Kernel::k_function(const PREFIX(node) *x, const PREFIX(node) *y,
 				else
 				{
 					if(x->index > y->index)
-					{
+					{	
 						sum += y->value * y->value;
 						++y;
 					}
@@ -524,7 +524,7 @@ double Kernel::k_function(const PREFIX(node) *x, const PREFIX(node) *y,
 #endif
                     }
 		default:
-			return 0;  // Unreachable
+			return 0;  // Unreachable 
 	}
 }
 // An SMO algorithm in Fan et al., JMLR 6(2005), p. 1889--1918
@@ -602,7 +602,7 @@ class Solver {
 	virtual double calculate_rho();
 	virtual void do_shrinking();
 private:
-	bool be_shrunk(int i, double Gmax1, double Gmax2);
+	bool be_shrunk(int i, double Gmax1, double Gmax2);	
 };
 
 void Solver::swap_index(int i, int j)
@@ -750,11 +750,11 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 			else
 				counter = 1;	// do shrinking next iteration
 		}
-
+		
 		++iter;
 
 		// update alpha[i] and alpha[j], handle bounds carefully
-
+		
 		const Qfloat *Q_i = Q.get_Q(i,active_size);
 		const Qfloat *Q_j = Q.get_Q(j,active_size);
 
@@ -773,7 +773,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 			double diff = alpha[i] - alpha[j];
 			alpha[i] += delta;
 			alpha[j] += delta;
-
+			
 			if(diff > 0)
 			{
 				if(alpha[j] < 0)
@@ -855,7 +855,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
 
 		double delta_alpha_i = alpha[i] - old_alpha_i;
 		double delta_alpha_j = alpha[j] - old_alpha_j;
-
+		
 		for(int k=0;k<active_size;k++)
 		{
 			G[k] += Q_i[k]*delta_alpha_i + Q_j[k]*delta_alpha_j;
@@ -947,7 +947,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 	// j: minimizes the decrease of obj value
 	//    (if quadratic coefficient <= 0, replace it with tau)
 	//    -y_j*grad(f)_j < -y_i*grad(f)_i, j in I_low(\alpha)
-
+	
 	double Gmax = -INF;
 	double Gmax2 = -INF;
 	int Gmax_idx = -1;
@@ -955,7 +955,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 	double obj_diff_min = INF;
 
 	for(int t=0;t<active_size;t++)
-		if(y[t]==+1)
+		if(y[t]==+1)	
 		{
 			if(!is_upper_bound(t))
 				if(-G[t] >= Gmax)
@@ -990,7 +990,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 					Gmax2 = G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff;
+					double obj_diff; 
 					double quad_coef = QD[i]+QD[j]-2.0*y[i]*Q_i[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1014,7 +1014,7 @@ int Solver::select_working_set(int &out_i, int &out_j)
 					Gmax2 = -G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff;
+					double obj_diff; 
 					double quad_coef = QD[i]+QD[j]+2.0*y[i]*Q_i[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1052,7 +1052,7 @@ bool Solver::be_shrunk(int i, double Gmax1, double Gmax2)
 	{
 		if(y[i]==+1)
 			return(G[i] > Gmax2);
-		else
+		else	
 			return(G[i] > Gmax1);
 	}
 	else
@@ -1068,27 +1068,27 @@ void Solver::do_shrinking()
 	// find maximal violating pair first
 	for(i=0;i<active_size;i++)
 	{
-		if(y[i]==+1)
+		if(y[i]==+1)	
 		{
-			if(!is_upper_bound(i))
+			if(!is_upper_bound(i))	
 			{
 				if(-G[i] >= Gmax1)
 					Gmax1 = -G[i];
 			}
-			if(!is_lower_bound(i))
+			if(!is_lower_bound(i))	
 			{
 				if(G[i] >= Gmax2)
 					Gmax2 = G[i];
 			}
 		}
-		else
+		else	
 		{
-			if(!is_upper_bound(i))
+			if(!is_upper_bound(i))	
 			{
 				if(-G[i] >= Gmax2)
 					Gmax2 = -G[i];
 			}
-			if(!is_lower_bound(i))
+			if(!is_lower_bound(i))	
 			{
 				if(G[i] >= Gmax1)
 					Gmax1 = G[i];
@@ -1096,7 +1096,7 @@ void Solver::do_shrinking()
 		}
 	}
 
-	if(unshrink == false && Gmax1 + Gmax2 <= eps*10)
+	if(unshrink == false && Gmax1 + Gmax2 <= eps*10) 
 	{
 		unshrink = true;
 		reconstruct_gradient();
@@ -1235,14 +1235,14 @@ int Solver_NU::select_working_set(int &out_i, int &out_j)
 	{
 		if(y[j]==+1)
 		{
-			if (!is_lower_bound(j))
+			if (!is_lower_bound(j))	
 			{
 				double grad_diff=Gmaxp+G[j];
 				if (G[j] >= Gmaxp2)
 					Gmaxp2 = G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff;
+					double obj_diff; 
 					double quad_coef = QD[ip]+QD[j]-2*Q_ip[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1266,7 +1266,7 @@ int Solver_NU::select_working_set(int &out_i, int &out_j)
 					Gmaxn2 = -G[j];
 				if (grad_diff > 0)
 				{
-					double obj_diff;
+					double obj_diff; 
 					double quad_coef = QD[in]+QD[j]-2*Q_in[j];
 					if (quad_coef > 0)
 						obj_diff = -(grad_diff*grad_diff)/quad_coef;
@@ -1301,14 +1301,14 @@ bool Solver_NU::be_shrunk(int i, double Gmax1, double Gmax2, double Gmax3, doubl
 	{
 		if(y[i]==+1)
 			return(-G[i] > Gmax1);
-		else
+		else	
 			return(-G[i] > Gmax4);
 	}
 	else if(is_lower_bound(i))
 	{
 		if(y[i]==+1)
 			return(G[i] > Gmax2);
-		else
+		else	
 			return(G[i] > Gmax3);
 	}
 	else
@@ -1337,14 +1337,14 @@ void Solver_NU::do_shrinking()
 		if(!is_lower_bound(i))
 		{
 			if(y[i]==+1)
-			{
+			{	
 				if(G[i] > Gmax2) Gmax2 = G[i];
 			}
 			else	if(G[i] > Gmax3) Gmax3 = G[i];
 		}
 	}
 
-	if(unshrink == false && max(Gmax1+Gmax2,Gmax3+Gmax4) <= eps*10)
+	if(unshrink == false && max(Gmax1+Gmax2,Gmax3+Gmax4) <= eps*10) 
 	{
 		unshrink = true;
 		reconstruct_gradient();
@@ -1407,12 +1407,12 @@ double Solver_NU::calculate_rho()
 		r1 = sum_free1/nr_free1;
 	else
 		r1 = (ub1+lb1)/2;
-
+	
 	if(nr_free2 > 0)
 		r2 = sum_free2/nr_free2;
 	else
 		r2 = (ub2+lb2)/2;
-
+	
 	si->r = (r1+r2)/2;
 	return (r1-r2)/2;
 }
@@ -1421,7 +1421,7 @@ double Solver_NU::calculate_rho()
 // Q matrices for various formulations
 //
 class SVC_Q: public Kernel
-{
+{ 
 public:
 	SVC_Q(const PREFIX(problem)& prob, const svm_parameter& param, const schar *y_, BlasFunctions *blas_functions)
 	:Kernel(prob.l, prob.x, param, blas_functions)
@@ -1432,7 +1432,7 @@ class SVC_Q: public Kernel
 		for(int i=0;i<prob.l;i++)
 			QD[i] = (this->*kernel_function)(i,i);
 	}
-
+	
 	Qfloat *get_Q(int i, int len) const
 	{
 		Qfloat *data;
@@ -1481,7 +1481,7 @@ class ONE_CLASS_Q: public Kernel
 		for(int i=0;i<prob.l;i++)
 			QD[i] = (this->*kernel_function)(i,i);
 	}
-
+	
 	Qfloat *get_Q(int i, int len) const
 	{
 		Qfloat *data;
@@ -1517,7 +1517,7 @@ class ONE_CLASS_Q: public Kernel
 };
 
 class SVR_Q: public Kernel
-{
+{ 
 public:
 	SVR_Q(const PREFIX(problem)& prob, const svm_parameter& param, BlasFunctions *blas_functions)
 	:Kernel(prob.l, prob.x, param, blas_functions)
@@ -1547,7 +1547,7 @@ class SVR_Q: public Kernel
 		swap(index[i],index[j]);
 		swap(QD[i],QD[j]);
 	}
-
+	
 	Qfloat *get_Q(int i, int len) const
 	{
 		Qfloat *data;
@@ -1663,7 +1663,7 @@ static void solve_nu_svc(
 
 		C[i] = prob->W[i];
 	}
-
+	
 	double nu_l = 0;
 	for(i=0;i<l;i++) nu_l += nu*C[i];
 	double sum_pos = nu_l/2;
@@ -1696,7 +1696,7 @@ static void solve_nu_svc(
 	for(i=0;i<l;i++)
         {
 		alpha[i] *= y[i]/r;
-		si->upper_bound[i] /= r;
+		si->upper_bound[i] /= r;                
         }
 
 	si->rho /= r;
@@ -1844,7 +1844,7 @@ static void solve_nu_svr(
 struct decision_function
 {
 	double *alpha;
-	double rho;
+	double rho;	
 	int n_iter;
 };
 
@@ -1857,23 +1857,23 @@ static decision_function svm_train_one(
 	switch(param->svm_type)
 	{
  		case C_SVC:
-			si.upper_bound = Malloc(double,prob->l);
+			si.upper_bound = Malloc(double,prob->l); 
  			solve_c_svc(prob,param,alpha,&si,Cp,Cn,blas_functions);
  			break;
  		case NU_SVC:
-			si.upper_bound = Malloc(double,prob->l);
+			si.upper_bound = Malloc(double,prob->l); 
  			solve_nu_svc(prob,param,alpha,&si,blas_functions);
  			break;
  		case ONE_CLASS:
-			si.upper_bound = Malloc(double,prob->l);
+			si.upper_bound = Malloc(double,prob->l); 
  			solve_one_class(prob,param,alpha,&si,blas_functions);
  			break;
  		case EPSILON_SVR:
-			si.upper_bound = Malloc(double,2*prob->l);
+			si.upper_bound = Malloc(double,2*prob->l); 
  			solve_epsilon_svr(prob,param,alpha,&si,blas_functions);
  			break;
  		case NU_SVR:
-			si.upper_bound = Malloc(double,2*prob->l);
+			si.upper_bound = Malloc(double,2*prob->l); 
  			solve_nu_svr(prob,param,alpha,&si,blas_functions);
  			break;
 	}
@@ -1917,7 +1917,7 @@ static decision_function svm_train_one(
 
 // Platt's binary SVM Probabilistic Output: an improvement from Lin et al.
 static void sigmoid_train(
-	int l, const double *dec_values, const double *labels,
+	int l, const double *dec_values, const double *labels, 
 	double& A, double& B)
 {
 	double prior1=0, prior0 = 0;
@@ -1926,7 +1926,7 @@ static void sigmoid_train(
 	for (i=0;i<l;i++)
 		if (labels[i] > 0) prior1+=1;
 		else prior0+=1;
-
+	
 	int max_iter=100;	// Maximal number of iterations
 	double min_step=1e-10;	// Minimal step taken in line search
 	double sigma=1e-12;	// For numerically strict PD of Hessian
@@ -1936,8 +1936,8 @@ static void sigmoid_train(
 	double *t=Malloc(double,l);
 	double fApB,p,q,h11,h22,h21,g1,g2,det,dA,dB,gd,stepsize;
 	double newA,newB,newf,d1,d2;
-	int iter;
-
+	int iter; 
+	
 	// Initial Point and Initial Fun Value
 	A=0.0; B=log((prior0+1.0)/(prior1+1.0));
 	double fval = 0.0;
@@ -2047,7 +2047,7 @@ static void multiclass_probability(int k, double **r, double *p)
 	double **Q=Malloc(double *,k);
 	double *Qp=Malloc(double,k);
 	double pQp, eps=0.005/k;
-
+	
 	for (t=0;t<k;t++)
 	{
 		p[t]=1.0/k;  // Valid if k = 1
@@ -2083,7 +2083,7 @@ static void multiclass_probability(int k, double **r, double *p)
 				max_error=error;
 		}
 		if (max_error<eps) break;
-
+		
 		for (t=0;t<k;t++)
 		{
 			double diff=(-Qp[t]+pQp)/Q[t][t];
@@ -2135,7 +2135,7 @@ static void svm_binary_svc_probability(
 #endif
 		subprob.y = Malloc(double,subprob.l);
                 subprob.W = Malloc(double,subprob.l);
-
+			
 		k=0;
 		for(j=0;j<begin;j++)
 		{
@@ -2183,26 +2183,26 @@ static void svm_binary_svc_probability(
 			for(j=begin;j<end;j++)
 			{
 #ifdef _DENSE_REP
-                                PREFIX(predict_values)(submodel,(prob->x+perm[j]),&(dec_values[perm[j]]), blas_functions);
+                                PREFIX(predict_values)(submodel,(prob->x+perm[j]),&(dec_values[perm[j]]), blas_functions); 
 #else
-				PREFIX(predict_values)(submodel,prob->x[perm[j]],&(dec_values[perm[j]]), blas_functions);
+				PREFIX(predict_values)(submodel,prob->x[perm[j]],&(dec_values[perm[j]]), blas_functions); 
 #endif
 				// ensure +1 -1 order; reason not using CV subroutine
 				dec_values[perm[j]] *= submodel->label[0];
-			}
+			}		
 			PREFIX(free_and_destroy_model)(&submodel);
 			PREFIX(destroy_param)(&subparam);
 		}
 		free(subprob.x);
 		free(subprob.y);
                 free(subprob.W);
-	}
+	}		
 	sigmoid_train(prob->l,dec_values,prob->y,probA,probB);
 	free(dec_values);
 	free(perm);
 }
 
-// Return parameter of a Laplace distribution
+// Return parameter of a Laplace distribution 
 static double svm_svr_probability(
 	const PREFIX(problem) *prob, const svm_parameter *param, BlasFunctions *blas_functions)
 {
@@ -2220,15 +2220,15 @@ static double svm_svr_probability(
 	{
 		ymv[i]=prob->y[i]-ymv[i];
 		mae += fabs(ymv[i]);
-	}
+	}		
 	mae /= prob->l;
 	double std=sqrt(2*mae*mae);
 	int count=0;
 	mae=0;
 	for(i=0;i<prob->l;i++)
-		if (fabs(ymv[i]) > 5*std)
+		if (fabs(ymv[i]) > 5*std) 
 			count=count+1;
-		else
+		else 
 			mae+=fabs(ymv[i]);
 	mae /= (prob->l-count);
 	info("Prob. model for test data: target value = predicted value + z,\nz: Laplace distribution e^(-|z|/sigma)/(2sigma),sigma= %g\n",mae);
@@ -2247,7 +2247,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
 	int nr_class = 0;
 	int *label = Malloc(int,max_nr_class);
 	int *count = Malloc(int,max_nr_class);
-	int *data_label = Malloc(int,l);
+	int *data_label = Malloc(int,l);	
 	int i, j, this_label, this_count;
 
 	for(i=0;i<l;i++)
@@ -2275,7 +2275,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
 		}
 	}
 
-        /*
+        /* 
          * Sort labels by straight insertion and apply the same
          * transformation to array count.
          */
@@ -2302,7 +2302,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
                         j ++;
                 }
                 data_label[i] = j;
-        }
+        }                
 
 	int *start = Malloc(int,nr_class);
 	start[0] = 0;
@@ -2329,7 +2329,7 @@ static void svm_group_classes(const PREFIX(problem) *prob, int *nr_class_ret, in
 
 // Remove zero weighed data as libsvm and some liblinear solvers require C > 0.
 //
-static void remove_zero_weight(PREFIX(problem) *newprob, const PREFIX(problem) *prob)
+static void remove_zero_weight(PREFIX(problem) *newprob, const PREFIX(problem) *prob) 
 {
 	int i;
 	int l = 0;
@@ -2386,7 +2386,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		model->probA = NULL; model->probB = NULL;
 		model->sv_coef = Malloc(double *,1);
 
-		if(param->probability &&
+		if(param->probability && 
 		   (param->svm_type == EPSILON_SVR ||
 		    param->svm_type == NU_SVR))
 		{
@@ -2420,7 +2420,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
                                 model->sv_ind[j] = i;
 				model->sv_coef[0][j] = f.alpha[i];
 				++j;
-			}
+			}		
 
 		free(f.alpha);
 	}
@@ -2435,7 +2435,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		int *perm = Malloc(int,l);
 
 		// group training data of the same class
-                NAMESPACE::svm_group_classes(prob,&nr_class,&label,&start,&count,perm);
+                NAMESPACE::svm_group_classes(prob,&nr_class,&label,&start,&count,perm);		
 #ifdef _DENSE_REP
 		PREFIX(node) *x = Malloc(PREFIX(node),l);
 #else
@@ -2456,7 +2456,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		for(i=0;i<nr_class;i++)
 			weighted_C[i] = param->C;
 		for(i=0;i<param->nr_weight;i++)
-		{
+		{	
 			int j;
 			for(j=0;j<nr_class;j++)
 				if(param->weight_label[i] == label[j])
@@ -2468,7 +2468,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		}
 
 		// train k*(k-1)/2 models
-
+		
 		bool *nonzero = Malloc(bool,l);
 		for(i=0;i<l;i++)
 			nonzero[i] = false;
@@ -2529,11 +2529,11 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 		// build output
 
 		model->nr_class = nr_class;
-
+		
 		model->label = Malloc(int,nr_class);
 		for(i=0;i<nr_class;i++)
 			model->label[i] = label[i];
-
+		
 		model->rho = Malloc(double,nr_class*(nr_class-1)/2);
 		model->n_iter = Malloc(int,nr_class*(nr_class-1)/2);
 		for(i=0;i<nr_class*(nr_class-1)/2;i++)
@@ -2566,7 +2566,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 			int nSV = 0;
 			for(int j=0;j<count[i];j++)
 				if(nonzero[start[i]+j])
-				{
+				{	
 					++nSV;
 					++total_sv;
 				}
@@ -2585,7 +2585,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 #endif
 		p = 0;
 		for(i=0;i<l;i++) {
-			if(nonzero[i]) {
+			if(nonzero[i]) { 
                                 model->SV[p] = x[i];
                                 model->sv_ind[p] = perm[i];
                                 ++p;
@@ -2613,7 +2613,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 				int sj = start[j];
 				int ci = count[i];
 				int cj = count[j];
-
+				
 				int q = nz_start[i];
 				int k;
 				for(k=0;k<ci;k++)
@@ -2625,7 +2625,7 @@ PREFIX(model) *PREFIX(train)(const PREFIX(problem) *prob, const svm_parameter *p
 						model->sv_coef[i][q++] = f[p].alpha[ci+k];
 				++p;
 			}
-
+		
 		free(label);
 		free(probA);
 		free(probB);
@@ -2677,7 +2677,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		int *index = Malloc(int,l);
 		for(i=0;i<l;i++)
 			index[i]=perm[i];
-		for (c=0; c<nr_class; c++)
+		for (c=0; c<nr_class; c++) 
 			for(i=0;i<count[c];i++)
 			{
 				int j = i+bounded_rand_int(count[c]-i);
@@ -2706,9 +2706,9 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		fold_start[0]=0;
 		for (i=1;i<=nr_fold;i++)
 			fold_start[i] = fold_start[i-1]+fold_count[i-1];
-		free(start);
+		free(start);	
 		free(label);
-		free(count);
+		free(count);	
 		free(index);
 		free(fold_count);
 	}
@@ -2739,7 +2739,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 #endif
 		subprob.y = Malloc(double,subprob.l);
 		subprob.W = Malloc(double,subprob.l);
-
+			
 		k=0;
 		for(j=0;j<begin;j++)
 		{
@@ -2757,7 +2757,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		}
                 int dummy_status = 0; // IGNORES TIMEOUT ERRORS
 		struct PREFIX(model) *submodel = PREFIX(train)(&subprob,param, &dummy_status, blas_functions);
-		if(param->probability &&
+		if(param->probability && 
 		   (param->svm_type == C_SVC || param->svm_type == NU_SVC))
 		{
 			double *prob_estimates=Malloc(double, PREFIX(get_nr_class)(submodel));
@@ -2767,7 +2767,7 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 #else
                                 target[perm[j]] = PREFIX(predict_probability)(submodel,prob->x[perm[j]],prob_estimates, blas_functions);
 #endif
-			free(prob_estimates);
+			free(prob_estimates);			
 		}
 		else
 			for(j=begin;j<end;j++)
@@ -2780,9 +2780,9 @@ void PREFIX(cross_validation)(const PREFIX(problem) *prob, const svm_parameter *
 		free(subprob.x);
 		free(subprob.y);
                 free(subprob.W);
-	}
+	}		
 	free(fold_start);
-	free(perm);
+	free(perm);	
 }
 
 
@@ -2824,7 +2824,7 @@ double PREFIX(predict_values)(const PREFIX(model) *model, const PREFIX(node) *x,
 	{
 		double *sv_coef = model->sv_coef[0];
 		double sum = 0;
-
+		
 		for(i=0;i<model->l;i++)
 #ifdef _DENSE_REP
                     sum += sv_coef[i] * NAMESPACE::Kernel::k_function(x,model->SV+i,model->param,blas_functions);
@@ -2843,7 +2843,7 @@ double PREFIX(predict_values)(const PREFIX(model) *model, const PREFIX(node) *x,
 	{
 		int nr_class = model->nr_class;
 		int l = model->l;
-
+		
 		double *kvalue = Malloc(double,l);
 		for(i=0;i<l;i++)
 #ifdef _DENSE_REP
@@ -2870,7 +2870,7 @@ double PREFIX(predict_values)(const PREFIX(model) *model, const PREFIX(node) *x,
 				int sj = start[j];
 				int ci = model->nSV[i];
 				int cj = model->nSV[j];
-
+				
 				int k;
 				double *coef1 = model->sv_coef[j-1];
 				double *coef2 = model->sv_coef[i];
@@ -2908,7 +2908,7 @@ double PREFIX(predict)(const PREFIX(model) *model, const PREFIX(node) *x, BlasFu
 	   model->param.svm_type == EPSILON_SVR ||
 	   model->param.svm_type == NU_SVR)
 		dec_values = Malloc(double, 1);
-	else
+	else 
 		dec_values = Malloc(double, nr_class*(nr_class-1)/2);
 	double pred_result = PREFIX(predict_values)(model, x, dec_values, blas_functions);
 	free(dec_values);
@@ -2947,10 +2947,10 @@ double PREFIX(predict_probability)(
 		for(i=0;i<nr_class;i++)
 			free(pairwise_prob[i]);
 		free(dec_values);
-		free(pairwise_prob);
+		free(pairwise_prob);	     
 		return model->label[prob_max_idx];
 	}
-	else
+	else 
 		return PREFIX(predict)(model, x, blas_functions);
 }
 
@@ -3026,9 +3026,9 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 	   svm_type != EPSILON_SVR &&
 	   svm_type != NU_SVR)
 		return "unknown svm type";
-
+	
 	// kernel_type, degree
-
+	
 	int kernel_type = param->kernel_type;
 	if(kernel_type != LINEAR &&
 	   kernel_type != POLY &&
@@ -3081,7 +3081,7 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 
 
 	// check whether nu-svc is feasible
-
+	
 	if(svm_type == NU_SVC)
 	{
 		int l = prob->l;
@@ -3115,7 +3115,7 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 				++nr_class;
 			}
 		}
-
+	
 		for(i=0;i<nr_class;i++)
 		{
 			double n1 = count[i];
@@ -3140,14 +3140,14 @@ const char *PREFIX(check_parameter)(const PREFIX(problem) *prob, const svm_param
 	   svm_type == ONE_CLASS)
 	{
 		PREFIX(problem) newprob;
-		// filter samples with negative and null weights
+		// filter samples with negative and null weights 
 		remove_zero_weight(&newprob, prob);
 
 		char* msg = NULL;
 		// all samples were removed
 		if(newprob.l == 0)
 			msg =  "Invalid input - all samples have zero or negative weights.";
-		else if(prob->l != newprob.l &&
+		else if(prob->l != newprob.l && 
 		        svm_type == C_SVC)
 		{
 			bool only_one_label = true;

From 026935b2703bc99ab8c604dcb0f83fce4f7da296 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Wed, 27 Oct 2021 12:33:32 -0300
Subject: [PATCH 11/28] Change the type of numpy arrays

---
 sklearn/svm/_libsvm.pyx        | 10 +++++-----
 sklearn/svm/_libsvm_sparse.pyx | 10 +++++-----
 2 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index aa55e6ba3a3d2..cbe1fe9893564 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -205,8 +205,8 @@ def fit(
     SV_len  = get_l(model)
     n_class = get_nr(model)
 
-    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter
-    n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
+    cdef np.ndarray[int, ndim=1, mode='c'] n_iter
+    n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.intc)
     copy_n_iter(n_iter.data, model)
 
     cdef np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef
@@ -289,7 +289,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
             np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
             np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-            np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter,
+            np.ndarray[int, ndim=1, mode='c'] n_iter,
             np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
             np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
             int svm_type=0, kernel='rbf', int degree=3,
@@ -388,7 +388,7 @@ def predict_proba(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter,
+    np.ndarray[int, ndim=1, mode='c'] n_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -497,7 +497,7 @@ def decision_function(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter,
+    np.ndarray[int, ndim=1, mode='c'] n_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
diff --git a/sklearn/svm/_libsvm_sparse.pyx b/sklearn/svm/_libsvm_sparse.pyx
index f21e9daa6d962..b73c2cd487a4b 100644
--- a/sklearn/svm/_libsvm_sparse.pyx
+++ b/sklearn/svm/_libsvm_sparse.pyx
@@ -160,8 +160,8 @@ def libsvm_sparse_train ( int n_features,
     cdef np.npy_intp SV_len = get_l(model)
     cdef np.npy_intp n_class = get_nr(model)
 
-    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter
-    n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.int32)
+    cdef np.ndarray[int, ndim=1, mode='c'] n_iter
+    n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.intc)
     copy_n_iter(n_iter.data, model)
 
     # copy model.sv_coef
@@ -242,7 +242,7 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
                             np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
                             np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-                            np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter):
+                            np.ndarray[int, ndim=1, mode='c'] n_iter):
     """
     Predict values T given a model.
 
@@ -320,7 +320,7 @@ def libsvm_sparse_predict_proba(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter):
+    np.ndarray[int, ndim=1, mode='c'] n_iter):
     """
     Predict values T given a model.
     """
@@ -382,7 +382,7 @@ def libsvm_sparse_decision_function(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] n_iter):
+    np.ndarray[int, ndim=1, mode='c'] n_iter):
     """
     Predict margin (libsvm name for this is predict_values)
 

From 164932da79337527740c9a0cfb889ad315415f30 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Thu, 28 Oct 2021 11:07:31 -0300
Subject: [PATCH 12/28] Apply reviewers suggestions

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 sklearn/svm/src/libsvm/libsvm_helper.c        | 9 ++++++++-
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c | 8 +++++++-
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index b91cb8870d2c9..0114dcbb0a751 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -2,6 +2,13 @@
 #include <numpy/arrayobject.h>
 #include "svm.h"
 #include "_svm_cython_blas_helpers.h"
+
+
+#ifndef max
+    #define max(x, y) (((x) > (y)) ? (x) : (y))
+#endif
+
+
 /*
  * Some helper methods for libsvm bindings.
  *
@@ -229,7 +236,7 @@ npy_intp get_nr(struct svm_model *model)
 void copy_n_iter(char *data, struct svm_model *model)
 {
     int n_models;
-    n_models = model->nr_class <= 2 ? 1 : model->nr_class * (model->nr_class-1)/2;
+    n_models = max(1, model->nr_class * (model->nr_class-1)/2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index 2779bfbfeed73..7e6bf6a9206fa 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -3,6 +3,12 @@
 #include "svm.h"
 #include "_svm_cython_blas_helpers.h"
 
+
+#ifndef max
+    #define max(x, y) (((x) > (y)) ? (x) : (y))
+#endif
+
+
 /*
  * Convert scipy.sparse.csr to libsvm's sparse data structure
  */
@@ -359,7 +365,7 @@ void copy_sv_coef(char *data, struct svm_csr_model *model)
 void copy_n_iter(char *data, struct svm_csr_model *model)
 {
     int n_models;
-    n_models = model->nr_class <= 2 ? 1 : model->nr_class * (model->nr_class-1)/2;
+    n_models = max(1, model->nr_class * (model->nr_class-1)/2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 

From d0fb264081b8dd2b12602d0baf38da1fea825041 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Thu, 28 Oct 2021 11:40:50 -0300
Subject: [PATCH 13/28] Apply reviewers suggestions

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 sklearn/svm/src/libsvm/libsvm_helper.c        | 3 +--
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index 0114dcbb0a751..c11ff6703fd77 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -235,8 +235,7 @@ npy_intp get_nr(struct svm_model *model)
  */
 void copy_n_iter(char *data, struct svm_model *model)
 {
-    int n_models;
-    n_models = max(1, model->nr_class * (model->nr_class-1)/2);
+    const int n_models = max(1, model->nr_class * (model->nr_class-1)/2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index 7e6bf6a9206fa..e94dc350ff3c2 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -364,8 +364,7 @@ void copy_sv_coef(char *data, struct svm_csr_model *model)
  */
 void copy_n_iter(char *data, struct svm_csr_model *model)
 {
-    int n_models;
-    n_models = max(1, model->nr_class * (model->nr_class-1)/2);
+    const int n_models = max(1, model->nr_class * (model->nr_class-1)/2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 

From 7a61d14924403b010df63498fd26ef201646dfb4 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Fri, 29 Oct 2021 23:50:05 -0300
Subject: [PATCH 14/28] Add FittedSVMAttributes class to improve code

Apply reviewers suggestions
Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 sklearn/svm/_base.py          |  70 ++++++++++----------
 sklearn/svm/_libsvm.pyx       | 116 +++++++++++++++++++++++-----------
 sklearn/svm/tests/test_svm.py |  52 ++-------------
 3 files changed, 120 insertions(+), 118 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 614447e4467a1..4b36eb236e600 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -136,6 +136,32 @@ def _pairwise(self):
         # Used by cross_val_score.
         return self.kernel == "precomputed"
 
+    def _set_fitted_attributes(self, fitted_attributes):
+        self.support_ = fitted_attributes.support
+        self.support_vectors_ = fitted_attributes.support_vectors
+        self._n_support = fitted_attributes.n_class_SV
+        self.dual_coef_ = fitted_attributes.sv_coef
+        self.intercept_ = fitted_attributes.intercept
+        self._probA = fitted_attributes.probA
+        self._probB = fitted_attributes.probB
+        self.fit_status_ = fitted_attributes.fit_status
+        self._num_iter = fitted_attributes.n_iter
+
+    def _get_fitted_attributes(self):
+        return libsvm.FittedSVMAttributes(
+            self.support_,
+            self.support_vectors_,
+            self._n_support,
+            # Note that we use the private attributes, since they are modified
+            # in the binary classification case
+            self._dual_coef_,
+            self._intercept_,
+            self._probA,
+            self._probB,
+            self.fit_status_,
+            self._num_iter,
+        )
+
     def fit(self, X, y, sample_weight=None):
         """Fit the SVM model according to the given training data.
 
@@ -312,17 +338,7 @@ def _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
 
         # we don't pass **self.get_params() to allow subclasses to
         # add other parameters to __init__
-        (
-            self.support_,
-            self.support_vectors_,
-            self._n_support,
-            self.dual_coef_,
-            self.intercept_,
-            self._probA,
-            self._probB,
-            self.fit_status_,
-            self._num_iter,
-        ) = libsvm.fit(
+        fitted_attributes = libsvm.fit(
             X,
             y,
             svm_type=solver_type,
@@ -342,6 +358,7 @@ def _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
             max_iter=self.max_iter,
             random_seed=random_seed,
         )
+        self._set_fitted_attributes(fitted_attributes)
 
         self._warn_from_fit_status()
 
@@ -442,17 +459,11 @@ def _dense_predict(self, X):
                 )
 
         svm_type = LIBSVM_IMPL.index(self._impl)
+        fitted_attributes = self._get_fitted_attributes()
 
         return libsvm.predict(
             X,
-            self.support_,
-            self.support_vectors_,
-            self._n_support,
-            self._dual_coef_,
-            self._intercept_,
-            self._num_iter,
-            self._probA,
-            self._probB,
+            fitted_attributes,
             svm_type=svm_type,
             kernel=kernel,
             degree=self.degree,
@@ -546,16 +557,11 @@ def _dense_decision_function(self, X):
         if callable(kernel):
             kernel = "precomputed"
 
+        fitted_attributes = self._get_fitted_attributes()
+
         return libsvm.decision_function(
             X,
-            self.support_,
-            self.support_vectors_,
-            self._n_support,
-            self._dual_coef_,
-            self._intercept_,
-            self._num_iter,
-            self._probA,
-            self._probB,
+            fitted_attributes,
             svm_type=LIBSVM_IMPL.index(self._impl),
             kernel=kernel,
             degree=self.degree,
@@ -894,16 +900,10 @@ def _dense_predict_proba(self, X):
             kernel = "precomputed"
 
         svm_type = LIBSVM_IMPL.index(self._impl)
+        fitted_attributes = self._get_fitted_attributes()
         pprob = libsvm.predict_proba(
             X,
-            self.support_,
-            self.support_vectors_,
-            self._n_support,
-            self._dual_coef_,
-            self._intercept_,
-            self._num_iter,
-            self._probA,
-            self._probB,
+            fitted_attributes,
             svm_type=svm_type,
             kernel=kernel,
             degree=self.degree,
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index cbe1fe9893564..811c31180b189 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -50,6 +50,57 @@ np.import_array()
 LIBSVM_KERNEL_TYPES = ['linear', 'poly', 'rbf', 'sigmoid', 'precomputed']
 
 
+################################################################################
+# Wrapper classes
+cdef class FittedSVMAttributes:
+    """
+    Wrapper class to hold the attributes returned by the LibSVM fit function.
+    """
+
+    cdef readonly np.ndarray support
+    cdef readonly np.ndarray support_vectors
+    cdef readonly np.ndarray n_class_SV
+    cdef readonly np.ndarray sv_coef
+    cdef readonly np.ndarray intercept
+    cdef readonly np.ndarray probA
+    cdef readonly np.ndarray probB
+    cdef readonly int fit_status
+    cdef readonly np.ndarray n_iter
+
+    # Use cinit to initialize all arrays to empty: this will prevent memory
+    # errors and seg-faults in rare cases where __init__ is not called
+    def __cinit__(self):
+        self.support = np.empty(1, dtype=np.int32, order='C')
+        self.support_vectors = np.empty((1, 1), dtype=np.float64, order='C')
+        self.n_class_SV = np.empty(1, dtype=np.int32, order='C')
+        self.sv_coef = np.empty((1, 1), dtype=np.float64, order='C')
+        self.intercept = np.empty(1, dtype=np.float64, order='C')
+        self.probA = np.empty(1, dtype=np.float64, order='C')
+        self.probB = np.empty(1, dtype=np.float64, order='C')
+        self.n_iter = np.empty(1, dtype=np.intc, order='C')
+
+    def __init__(self,
+        np.ndarray[np.int32_t, ndim=1, mode='c'] support,
+        np.ndarray[np.float64_t, ndim=2, mode='c'] support_vectors,
+        np.ndarray[np.int32_t, ndim=1, mode='c'] n_class_SV,
+        np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
+        np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+        np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
+        np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
+        int fit_status,
+        np.ndarray[int, ndim=1, mode='c'] n_iter):
+
+        self.support = support
+        self.support_vectors = support_vectors
+        self.n_class_SV = n_class_SV
+        self.sv_coef = sv_coef
+        self.intercept = intercept
+        self.probA = probA
+        self.probB = probB
+        self.fit_status = fit_status
+        self.n_iter = n_iter
+
+
 ################################################################################
 # Wrapper functions
 
@@ -257,8 +308,8 @@ def fit(
     svm_free_and_destroy_model(&model)
     free(problem.x)
 
-    return (support, support_vectors, n_class_SV, sv_coef, intercept,
-           probA, probB, fit_status, n_iter)
+    return FittedSVMAttributes(support, support_vectors, n_class_SV, sv_coef,
+                               intercept, probA, probB, fit_status, n_iter)
 
 
 cdef void set_predict_params(
@@ -284,14 +335,7 @@ cdef void set_predict_params(
 
 
 def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
-            np.ndarray[np.int32_t, ndim=1, mode='c'] support,
-            np.ndarray[np.float64_t, ndim=2, mode='c'] SV,
-            np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
-            np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
-            np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-            np.ndarray[int, ndim=1, mode='c'] n_iter,
-            np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
-            np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
+            FittedSVMAttributes fitted_att,
             int svm_type=0, kernel='rbf', int degree=3,
             double gamma=0.1, double coef0=0.,
             np.ndarray[np.float64_t, ndim=1, mode='c']
@@ -362,10 +406,14 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     set_predict_params(&param, svm_type, kernel, degree, gamma, coef0,
                        cache_size, 0, <int>class_weight.shape[0],
                        class_weight_label.data, class_weight.data)
-    model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
-                      support.data, support.shape, sv_coef.strides,
-                      sv_coef.data, intercept.data, nSV.data, probA.data,
-                      probB.data, n_iter.data)
+    model = set_model(&param, <int> fitted_att.n_class_SV.shape[0],
+                      fitted_att.support_vectors.data,
+                      fitted_att.support_vectors.shape,
+                      fitted_att.support.data, fitted_att.support.shape,
+                      fitted_att.sv_coef.strides, fitted_att.sv_coef.data,
+                      fitted_att.intercept.data, fitted_att.n_class_SV.data,
+                      fitted_att.probA.data, fitted_att.probB.data,
+                      fitted_att.n_iter.data)
     cdef BlasFunctions blas_functions
     blas_functions.dot = _dot[double]
     #TODO: use check_model
@@ -383,14 +431,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
 
 def predict_proba(
     np.ndarray[np.float64_t, ndim=2, mode='c'] X,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] support,
-    np.ndarray[np.float64_t, ndim=2, mode='c'] SV,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
-    np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[int, ndim=1, mode='c'] n_iter,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
+    FittedSVMAttributes fitted_att,
     int svm_type=0, kernel='rbf', int degree=3,
     double gamma=0.1, double coef0=0.,
     np.ndarray[np.float64_t, ndim=1, mode='c']
@@ -470,10 +511,13 @@ def predict_proba(
     set_predict_params(&param, svm_type, kernel, degree, gamma, coef0,
                        cache_size, 1, <int>class_weight.shape[0],
                        class_weight_label.data, class_weight.data)
-    model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
-                      support.data, support.shape, sv_coef.strides,
-                      sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data, n_iter.data)
+    model = set_model(&param, <int> fitted_att.n_class_SV.shape[0],
+                      fitted_att.support_vectors.data,
+                      fitted_att.support_vectors.shape, fitted_att.support.data,
+                      fitted_att.support.shape, fitted_att.sv_coef.strides,
+                      fitted_att.sv_coef.data, fitted_att.intercept.data,
+                      fitted_att.n_class_SV.data, fitted_att.probA.data,
+                      fitted_att.probB.data, fitted_att.n_iter.data)
 
     cdef np.npy_intp n_class = get_nr(model)
     cdef BlasFunctions blas_functions
@@ -492,14 +536,7 @@ def predict_proba(
 
 def decision_function(
     np.ndarray[np.float64_t, ndim=2, mode='c'] X,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] support,
-    np.ndarray[np.float64_t, ndim=2, mode='c'] SV,
-    np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
-    np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[int, ndim=1, mode='c'] n_iter,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
+    FittedSVMAttributes fitted_att,
     int svm_type=0, kernel='rbf', int degree=3,
     double gamma=0.1, double coef0=0.,
     np.ndarray[np.float64_t, ndim=1, mode='c']
@@ -576,10 +613,13 @@ def decision_function(
                        cache_size, 0, <int>class_weight.shape[0],
                        class_weight_label.data, class_weight.data)
 
-    model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
-                      support.data, support.shape, sv_coef.strides,
-                      sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data, n_iter.data)
+    model = set_model(&param, <int> fitted_att.n_class_SV.shape[0],
+                      fitted_att.support_vectors.data,
+                      fitted_att.support_vectors.shape, fitted_att.support.data,
+                      fitted_att.support.shape, fitted_att.sv_coef.strides,
+                      fitted_att.sv_coef.data, fitted_att.intercept.data,
+                      fitted_att.n_class_SV.data, fitted_att.probA.data,
+                      fitted_att.probB.data, fitted_att.n_iter.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index 17067183c9197..18048cfbef106 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -66,54 +66,16 @@ def test_libsvm_iris():
     assert_array_equal(clf.classes_, np.sort(clf.classes_))
 
     # check also the low-level API
-    (
-        libsvm_support,
-        libsvm_support_vectors,
-        libsvm_n_class_SV,
-        libsvm_sv_coef,
-        libsvm_intercept,
-        libsvm_probA,
-        libsvm_probB,
-        libsvm_fit_status,
-        libsvm_n_iter,
-    ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
-
-    lib_svm_model = (
-        libsvm_support,
-        libsvm_support_vectors,
-        libsvm_n_class_SV,
-        libsvm_sv_coef,
-        libsvm_intercept,
-        libsvm_n_iter,
-        libsvm_probA,
-        libsvm_probB,
-    )
-    pred = _libsvm.predict(iris.data, *lib_svm_model)
+    fitted_attributes = _libsvm.fit(iris.data, iris.target.astype(np.float64))
+
+    pred = _libsvm.predict(iris.data, fitted_attributes)
     assert np.mean(pred == iris.target) > 0.95
 
-    (
-        libsvm_support,
-        libsvm_support_vectors,
-        libsvm_n_class_SV,
-        libsvm_sv_coef,
-        libsvm_intercept,
-        libsvm_probA,
-        libsvm_probB,
-        libsvm_fit_status,
-        libsvm_n_iter,
-    ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
-
-    lib_svm_model = (
-        libsvm_support,
-        libsvm_support_vectors,
-        libsvm_n_class_SV,
-        libsvm_sv_coef,
-        libsvm_intercept,
-        libsvm_n_iter,
-        libsvm_probA,
-        libsvm_probB,
+    fitted_attributes = _libsvm.fit(
+        iris.data, iris.target.astype(np.float64), kernel="linear"
     )
-    pred = _libsvm.predict(iris.data, *lib_svm_model, kernel="linear")
+
+    pred = _libsvm.predict(iris.data, fitted_attributes, kernel="linear")
     assert np.mean(pred == iris.target) > 0.95
 
     pred = _libsvm.cross_validation(

From 8802a74be6d911b4dc1aa30dcc3b8884b0d24be6 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Sat, 30 Oct 2021 14:49:27 -0300
Subject: [PATCH 15/28] Revert "Add FittedSVMAttributes class"

This reverts commit 7a61d14924403b010df63498fd26ef201646dfb4.
---
 sklearn/svm/_base.py          |  70 ++++++++++----------
 sklearn/svm/_libsvm.pyx       | 116 +++++++++++-----------------------
 sklearn/svm/tests/test_svm.py |  52 +++++++++++++--
 3 files changed, 118 insertions(+), 120 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 4b36eb236e600..614447e4467a1 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -136,32 +136,6 @@ def _pairwise(self):
         # Used by cross_val_score.
         return self.kernel == "precomputed"
 
-    def _set_fitted_attributes(self, fitted_attributes):
-        self.support_ = fitted_attributes.support
-        self.support_vectors_ = fitted_attributes.support_vectors
-        self._n_support = fitted_attributes.n_class_SV
-        self.dual_coef_ = fitted_attributes.sv_coef
-        self.intercept_ = fitted_attributes.intercept
-        self._probA = fitted_attributes.probA
-        self._probB = fitted_attributes.probB
-        self.fit_status_ = fitted_attributes.fit_status
-        self._num_iter = fitted_attributes.n_iter
-
-    def _get_fitted_attributes(self):
-        return libsvm.FittedSVMAttributes(
-            self.support_,
-            self.support_vectors_,
-            self._n_support,
-            # Note that we use the private attributes, since they are modified
-            # in the binary classification case
-            self._dual_coef_,
-            self._intercept_,
-            self._probA,
-            self._probB,
-            self.fit_status_,
-            self._num_iter,
-        )
-
     def fit(self, X, y, sample_weight=None):
         """Fit the SVM model according to the given training data.
 
@@ -338,7 +312,17 @@ def _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
 
         # we don't pass **self.get_params() to allow subclasses to
         # add other parameters to __init__
-        fitted_attributes = libsvm.fit(
+        (
+            self.support_,
+            self.support_vectors_,
+            self._n_support,
+            self.dual_coef_,
+            self.intercept_,
+            self._probA,
+            self._probB,
+            self.fit_status_,
+            self._num_iter,
+        ) = libsvm.fit(
             X,
             y,
             svm_type=solver_type,
@@ -358,7 +342,6 @@ def _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed):
             max_iter=self.max_iter,
             random_seed=random_seed,
         )
-        self._set_fitted_attributes(fitted_attributes)
 
         self._warn_from_fit_status()
 
@@ -459,11 +442,17 @@ def _dense_predict(self, X):
                 )
 
         svm_type = LIBSVM_IMPL.index(self._impl)
-        fitted_attributes = self._get_fitted_attributes()
 
         return libsvm.predict(
             X,
-            fitted_attributes,
+            self.support_,
+            self.support_vectors_,
+            self._n_support,
+            self._dual_coef_,
+            self._intercept_,
+            self._num_iter,
+            self._probA,
+            self._probB,
             svm_type=svm_type,
             kernel=kernel,
             degree=self.degree,
@@ -557,11 +546,16 @@ def _dense_decision_function(self, X):
         if callable(kernel):
             kernel = "precomputed"
 
-        fitted_attributes = self._get_fitted_attributes()
-
         return libsvm.decision_function(
             X,
-            fitted_attributes,
+            self.support_,
+            self.support_vectors_,
+            self._n_support,
+            self._dual_coef_,
+            self._intercept_,
+            self._num_iter,
+            self._probA,
+            self._probB,
             svm_type=LIBSVM_IMPL.index(self._impl),
             kernel=kernel,
             degree=self.degree,
@@ -900,10 +894,16 @@ def _dense_predict_proba(self, X):
             kernel = "precomputed"
 
         svm_type = LIBSVM_IMPL.index(self._impl)
-        fitted_attributes = self._get_fitted_attributes()
         pprob = libsvm.predict_proba(
             X,
-            fitted_attributes,
+            self.support_,
+            self.support_vectors_,
+            self._n_support,
+            self._dual_coef_,
+            self._intercept_,
+            self._num_iter,
+            self._probA,
+            self._probB,
             svm_type=svm_type,
             kernel=kernel,
             degree=self.degree,
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index 811c31180b189..cbe1fe9893564 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -50,57 +50,6 @@ np.import_array()
 LIBSVM_KERNEL_TYPES = ['linear', 'poly', 'rbf', 'sigmoid', 'precomputed']
 
 
-################################################################################
-# Wrapper classes
-cdef class FittedSVMAttributes:
-    """
-    Wrapper class to hold the attributes returned by the LibSVM fit function.
-    """
-
-    cdef readonly np.ndarray support
-    cdef readonly np.ndarray support_vectors
-    cdef readonly np.ndarray n_class_SV
-    cdef readonly np.ndarray sv_coef
-    cdef readonly np.ndarray intercept
-    cdef readonly np.ndarray probA
-    cdef readonly np.ndarray probB
-    cdef readonly int fit_status
-    cdef readonly np.ndarray n_iter
-
-    # Use cinit to initialize all arrays to empty: this will prevent memory
-    # errors and seg-faults in rare cases where __init__ is not called
-    def __cinit__(self):
-        self.support = np.empty(1, dtype=np.int32, order='C')
-        self.support_vectors = np.empty((1, 1), dtype=np.float64, order='C')
-        self.n_class_SV = np.empty(1, dtype=np.int32, order='C')
-        self.sv_coef = np.empty((1, 1), dtype=np.float64, order='C')
-        self.intercept = np.empty(1, dtype=np.float64, order='C')
-        self.probA = np.empty(1, dtype=np.float64, order='C')
-        self.probB = np.empty(1, dtype=np.float64, order='C')
-        self.n_iter = np.empty(1, dtype=np.intc, order='C')
-
-    def __init__(self,
-        np.ndarray[np.int32_t, ndim=1, mode='c'] support,
-        np.ndarray[np.float64_t, ndim=2, mode='c'] support_vectors,
-        np.ndarray[np.int32_t, ndim=1, mode='c'] n_class_SV,
-        np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
-        np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-        np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-        np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-        int fit_status,
-        np.ndarray[int, ndim=1, mode='c'] n_iter):
-
-        self.support = support
-        self.support_vectors = support_vectors
-        self.n_class_SV = n_class_SV
-        self.sv_coef = sv_coef
-        self.intercept = intercept
-        self.probA = probA
-        self.probB = probB
-        self.fit_status = fit_status
-        self.n_iter = n_iter
-
-
 ################################################################################
 # Wrapper functions
 
@@ -308,8 +257,8 @@ def fit(
     svm_free_and_destroy_model(&model)
     free(problem.x)
 
-    return FittedSVMAttributes(support, support_vectors, n_class_SV, sv_coef,
-                               intercept, probA, probB, fit_status, n_iter)
+    return (support, support_vectors, n_class_SV, sv_coef, intercept,
+           probA, probB, fit_status, n_iter)
 
 
 cdef void set_predict_params(
@@ -335,7 +284,14 @@ cdef void set_predict_params(
 
 
 def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
-            FittedSVMAttributes fitted_att,
+            np.ndarray[np.int32_t, ndim=1, mode='c'] support,
+            np.ndarray[np.float64_t, ndim=2, mode='c'] SV,
+            np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
+            np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
+            np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+            np.ndarray[int, ndim=1, mode='c'] n_iter,
+            np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
+            np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
             int svm_type=0, kernel='rbf', int degree=3,
             double gamma=0.1, double coef0=0.,
             np.ndarray[np.float64_t, ndim=1, mode='c']
@@ -406,14 +362,10 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     set_predict_params(&param, svm_type, kernel, degree, gamma, coef0,
                        cache_size, 0, <int>class_weight.shape[0],
                        class_weight_label.data, class_weight.data)
-    model = set_model(&param, <int> fitted_att.n_class_SV.shape[0],
-                      fitted_att.support_vectors.data,
-                      fitted_att.support_vectors.shape,
-                      fitted_att.support.data, fitted_att.support.shape,
-                      fitted_att.sv_coef.strides, fitted_att.sv_coef.data,
-                      fitted_att.intercept.data, fitted_att.n_class_SV.data,
-                      fitted_att.probA.data, fitted_att.probB.data,
-                      fitted_att.n_iter.data)
+    model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
+                      support.data, support.shape, sv_coef.strides,
+                      sv_coef.data, intercept.data, nSV.data, probA.data,
+                      probB.data, n_iter.data)
     cdef BlasFunctions blas_functions
     blas_functions.dot = _dot[double]
     #TODO: use check_model
@@ -431,7 +383,14 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
 
 def predict_proba(
     np.ndarray[np.float64_t, ndim=2, mode='c'] X,
-    FittedSVMAttributes fitted_att,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] support,
+    np.ndarray[np.float64_t, ndim=2, mode='c'] SV,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
+    np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
+    np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+    np.ndarray[int, ndim=1, mode='c'] n_iter,
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
     double gamma=0.1, double coef0=0.,
     np.ndarray[np.float64_t, ndim=1, mode='c']
@@ -511,13 +470,10 @@ def predict_proba(
     set_predict_params(&param, svm_type, kernel, degree, gamma, coef0,
                        cache_size, 1, <int>class_weight.shape[0],
                        class_weight_label.data, class_weight.data)
-    model = set_model(&param, <int> fitted_att.n_class_SV.shape[0],
-                      fitted_att.support_vectors.data,
-                      fitted_att.support_vectors.shape, fitted_att.support.data,
-                      fitted_att.support.shape, fitted_att.sv_coef.strides,
-                      fitted_att.sv_coef.data, fitted_att.intercept.data,
-                      fitted_att.n_class_SV.data, fitted_att.probA.data,
-                      fitted_att.probB.data, fitted_att.n_iter.data)
+    model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
+                      support.data, support.shape, sv_coef.strides,
+                      sv_coef.data, intercept.data, nSV.data,
+                      probA.data, probB.data, n_iter.data)
 
     cdef np.npy_intp n_class = get_nr(model)
     cdef BlasFunctions blas_functions
@@ -536,7 +492,14 @@ def predict_proba(
 
 def decision_function(
     np.ndarray[np.float64_t, ndim=2, mode='c'] X,
-    FittedSVMAttributes fitted_att,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] support,
+    np.ndarray[np.float64_t, ndim=2, mode='c'] SV,
+    np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
+    np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
+    np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
+    np.ndarray[int, ndim=1, mode='c'] n_iter,
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
     double gamma=0.1, double coef0=0.,
     np.ndarray[np.float64_t, ndim=1, mode='c']
@@ -613,13 +576,10 @@ def decision_function(
                        cache_size, 0, <int>class_weight.shape[0],
                        class_weight_label.data, class_weight.data)
 
-    model = set_model(&param, <int> fitted_att.n_class_SV.shape[0],
-                      fitted_att.support_vectors.data,
-                      fitted_att.support_vectors.shape, fitted_att.support.data,
-                      fitted_att.support.shape, fitted_att.sv_coef.strides,
-                      fitted_att.sv_coef.data, fitted_att.intercept.data,
-                      fitted_att.n_class_SV.data, fitted_att.probA.data,
-                      fitted_att.probB.data, fitted_att.n_iter.data)
+    model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
+                      support.data, support.shape, sv_coef.strides,
+                      sv_coef.data, intercept.data, nSV.data,
+                      probA.data, probB.data, n_iter.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index 18048cfbef106..17067183c9197 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -66,16 +66,54 @@ def test_libsvm_iris():
     assert_array_equal(clf.classes_, np.sort(clf.classes_))
 
     # check also the low-level API
-    fitted_attributes = _libsvm.fit(iris.data, iris.target.astype(np.float64))
-
-    pred = _libsvm.predict(iris.data, fitted_attributes)
+    (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_probA,
+        libsvm_probB,
+        libsvm_fit_status,
+        libsvm_n_iter,
+    ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
+
+    lib_svm_model = (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_n_iter,
+        libsvm_probA,
+        libsvm_probB,
+    )
+    pred = _libsvm.predict(iris.data, *lib_svm_model)
     assert np.mean(pred == iris.target) > 0.95
 
-    fitted_attributes = _libsvm.fit(
-        iris.data, iris.target.astype(np.float64), kernel="linear"
+    (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_probA,
+        libsvm_probB,
+        libsvm_fit_status,
+        libsvm_n_iter,
+    ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
+
+    lib_svm_model = (
+        libsvm_support,
+        libsvm_support_vectors,
+        libsvm_n_class_SV,
+        libsvm_sv_coef,
+        libsvm_intercept,
+        libsvm_n_iter,
+        libsvm_probA,
+        libsvm_probB,
     )
-
-    pred = _libsvm.predict(iris.data, fitted_attributes, kernel="linear")
+    pred = _libsvm.predict(iris.data, *lib_svm_model, kernel="linear")
     assert np.mean(pred == iris.target) > 0.95
 
     pred = _libsvm.cross_validation(

From fa626fb737185f37f7105d2cfa1595f00bb383ef Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Sat, 30 Oct 2021 15:05:03 -0300
Subject: [PATCH 16/28] Comment unpacking and repacking of values for test

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 sklearn/svm/tests/test_svm.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index 17067183c9197..a6b77a4ef98d6 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -66,6 +66,8 @@ def test_libsvm_iris():
     assert_array_equal(clf.classes_, np.sort(clf.classes_))
 
     # check also the low-level API
+    # We unpack the values to be able to repack some of the
+    # return values from libsvm.
     (
         libsvm_support,
         libsvm_support_vectors,
@@ -74,6 +76,7 @@ def test_libsvm_iris():
         libsvm_intercept,
         libsvm_probA,
         libsvm_probB,
+        # libsvm_fit_status won't be packed bellow.
         libsvm_fit_status,
         libsvm_n_iter,
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
@@ -91,6 +94,8 @@ def test_libsvm_iris():
     pred = _libsvm.predict(iris.data, *lib_svm_model)
     assert np.mean(pred == iris.target) > 0.95
 
+    # We unpack the values to be able to repack some of the
+    # return values from libsvm.
     (
         libsvm_support,
         libsvm_support_vectors,
@@ -99,6 +104,7 @@ def test_libsvm_iris():
         libsvm_intercept,
         libsvm_probA,
         libsvm_probB,
+        # libsvm_fit_status won't be packed bellow.
         libsvm_fit_status,
         libsvm_n_iter,
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")

From d1dc1ad80fbcc022d55a954257b59ab42ca73691 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 2 Nov 2021 15:44:51 -0300
Subject: [PATCH 17/28] Apply reviewers suggestions

- Improve `whats_new` message
- Use uppercase for macro name

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 doc/whats_new/v1.1.rst                        | 2 +-
 sklearn/svm/src/libsvm/libsvm_helper.c        | 6 +++---
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c | 6 +++---
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/doc/whats_new/v1.1.rst b/doc/whats_new/v1.1.rst
index dbda2178a3916..fc96d70ed079b 100644
--- a/doc/whats_new/v1.1.rst
+++ b/doc/whats_new/v1.1.rst
@@ -136,7 +136,7 @@ Changelog
 
 - |Enhancement| :class:`svm.OneClassSVM`, :class:`svm.NuSVC`,
   :class:`svm.NuSVR`, :class:`svm.SVC` and :class:`svm.SVR` now expose
-  `n_iter_`, the number of iterations used by the libsvm optimization routine.
+  `n_iter_`, the number of iterations of the libsvm optimization routine.
   :pr:`21408` by :user:`Juan Martín Loyola <jmloyola>`.
 :mod:`sklearn.preprocessing`
 ............................
diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index c11ff6703fd77..f9796f81e4b72 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -4,8 +4,8 @@
 #include "_svm_cython_blas_helpers.h"
 
 
-#ifndef max
-    #define max(x, y) (((x) > (y)) ? (x) : (y))
+#ifndef MAX
+    #define MAX(x, y) (((x) > (y)) ? (x) : (y))
 #endif
 
 
@@ -235,7 +235,7 @@ npy_intp get_nr(struct svm_model *model)
  */
 void copy_n_iter(char *data, struct svm_model *model)
 {
-    const int n_models = max(1, model->nr_class * (model->nr_class-1)/2);
+    const int n_models = MAX(1, model->nr_class * (model->nr_class-1)/2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index e94dc350ff3c2..5ad2106550027 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -4,8 +4,8 @@
 #include "_svm_cython_blas_helpers.h"
 
 
-#ifndef max
-    #define max(x, y) (((x) > (y)) ? (x) : (y))
+#ifndef MAX
+    #define MAX(x, y) (((x) > (y)) ? (x) : (y))
 #endif
 
 
@@ -364,7 +364,7 @@ void copy_sv_coef(char *data, struct svm_csr_model *model)
  */
 void copy_n_iter(char *data, struct svm_csr_model *model)
 {
-    const int n_models = max(1, model->nr_class * (model->nr_class-1)/2);
+    const int n_models = MAX(1, model->nr_class * (model->nr_class-1)/2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 

From d89a346fc21505ab787fe93d28eaa597f342860a Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 2 Nov 2021 16:30:16 -0300
Subject: [PATCH 18/28] Apply reviewers suggestions

- Fix check_transformer_n_iter test. It did not need a check for SVC or NuSVC since they are not transformers.
- Simplify the number of iterations assert.
- Use _pairwise_estimator_convert_X if using the precomputed kernel

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
---
 sklearn/utils/estimator_checks.py | 13 +++----------
 1 file changed, 3 insertions(+), 10 deletions(-)

diff --git a/sklearn/utils/estimator_checks.py b/sklearn/utils/estimator_checks.py
index 94d8010ec1e82..cda48da0bd279 100644
--- a/sklearn/utils/estimator_checks.py
+++ b/sklearn/utils/estimator_checks.py
@@ -3264,16 +3264,11 @@ def check_non_transformer_estimators_n_iter(name, estimator_orig):
 
         set_random_state(estimator, 0)
 
-        if name in ["SVC", "NuSVC"] and estimator.kernel == "precomputed":
-            X = np.dot(X, X.T)
+        X = _pairwise_estimator_convert_X(X, estimator_orig)
 
         estimator.fit(X, y_)
 
-        # These return a n_iter per model optimized
-        if name in ["SVC", "NuSVC"] and len(estimator.classes_) > 2:
-            assert np.all(estimator.n_iter_ >= 1)
-        else:
-            assert estimator.n_iter_ >= 1
+        assert np.all(estimator.n_iter_ >= 1)
 
 
 @ignore_warnings(category=FutureWarning)
@@ -3300,9 +3295,7 @@ def check_transformer_n_iter(name, estimator_orig):
         estimator.fit(X, y_)
 
         # These return a n_iter per component.
-        if (name in CROSS_DECOMPOSITION) or (
-            name in ["SVC", "NuSVC"] and len(estimator.classes_) > 2
-        ):
+        if name in CROSS_DECOMPOSITION:
             for iter_ in estimator.n_iter_:
                 assert iter_ >= 1
         else:

From eccafdcc38cf07bcc538980196916f70aa2d3553 Mon Sep 17 00:00:00 2001
From: Juan Martin Loyola <jmloyola@outlook.com>
Date: Tue, 2 Nov 2021 16:52:25 -0300
Subject: [PATCH 19/28] Update doc/whats_new/v1.1.rst

Co-authored-by: Julien Jerphanion <git@jjerphan.xyz>
---
 doc/whats_new/v1.1.rst | 1 +
 1 file changed, 1 insertion(+)

diff --git a/doc/whats_new/v1.1.rst b/doc/whats_new/v1.1.rst
index fc96d70ed079b..a8810ba069d73 100644
--- a/doc/whats_new/v1.1.rst
+++ b/doc/whats_new/v1.1.rst
@@ -138,6 +138,7 @@ Changelog
   :class:`svm.NuSVR`, :class:`svm.SVC` and :class:`svm.SVR` now expose
   `n_iter_`, the number of iterations of the libsvm optimization routine.
   :pr:`21408` by :user:`Juan Martín Loyola <jmloyola>`.
+
 :mod:`sklearn.preprocessing`
 ............................
 

From 87bbb0b30df63571e91413973e5fba72c38b1933 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Wed, 3 Nov 2021 13:00:21 -0300
Subject: [PATCH 20/28] Check n_iter for outlier_detector estimators with
 max_iter

---
 sklearn/utils/estimator_checks.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/sklearn/utils/estimator_checks.py b/sklearn/utils/estimator_checks.py
index cda48da0bd279..c57f8d78a0f68 100644
--- a/sklearn/utils/estimator_checks.py
+++ b/sklearn/utils/estimator_checks.py
@@ -277,6 +277,7 @@ def _yield_outliers_checks(estimator):
         # test if NotFittedError is raised
         if _safe_tags(estimator, key="requires_fit"):
             yield check_estimators_unfitted
+    yield check_non_transformer_estimators_n_iter
 
 
 def _yield_all_checks(estimator):

From 3728adec005f661998232c9baed04b49e1956d67 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Fri, 5 Nov 2021 17:55:19 -0300
Subject: [PATCH 21/28] Simplify the number of models check

Co-authored-by: Adrin Jalali <adrin.jalali@gmail.com>
---
 sklearn/svm/_base.py | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 614447e4467a1..1abad240683db 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -268,9 +268,7 @@ def fit(self, X, y, sample_weight=None):
 
         # If the number of models optimized by libSVM is one, get the number of
         # iterations as an integer instead of ndarray.
-        if (
-            self._impl in ["c_svc", "nu_svc"] and len(self.classes_) <= 2
-        ) or self._impl in ["one_class", "epsilon_svr", "nu_svr"]:
+        if len(self._num_iter) == 1:
             self.n_iter_ = self._num_iter[0]
         else:
             self.n_iter_ = self._num_iter

From d310c5a21c97418e20745e205a22dadf125036ab Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Sat, 6 Nov 2021 20:07:47 -0300
Subject: [PATCH 22/28] Change n_iter_ to always be ndarray for SVC and NuSVC

---
 sklearn/svm/_base.py    | 14 +++++++++-----
 sklearn/svm/_classes.py | 24 ++++++++++++++----------
 2 files changed, 23 insertions(+), 15 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 23bab337dfff0..0195b033284ed 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -274,12 +274,16 @@ def fit(self, X, y, sample_weight=None):
             self.intercept_ *= -1
             self.dual_coef_ = -self.dual_coef_
 
-        # If the number of models optimized by libSVM is one, get the number of
-        # iterations as an integer instead of ndarray.
-        if len(self._num_iter) == 1:
-            self.n_iter_ = self._num_iter[0]
-        else:
+        # Since, in the case of SVC and NuSVC, the number of models optimized by
+        # libSVM could be greater than one (depending on the input), `n_iter_`
+        # stores an ndarray.
+        # For the other sub-classes (SVR, NuSVR, and OneClassSVM), the number of
+        # models optimized by libSVM is always one, so `n_iter_` stores an
+        # integer.
+        if self._impl in ["c_svc", "nu_svc"]:
             self.n_iter_ = self._num_iter
+        else:
+            self.n_iter_ = self._num_iter[0]
 
         return self
 
diff --git a/sklearn/svm/_classes.py b/sklearn/svm/_classes.py
index 4f4bce25f88c3..fcd0f43984c9e 100644
--- a/sklearn/svm/_classes.py
+++ b/sklearn/svm/_classes.py
@@ -670,12 +670,14 @@ class SVC(BaseSVC):
 
         .. versionadded:: 1.0
 
-    n_iter_ : int or ndarray of shape (n_class * (n_class - 1) // 2,)
+    n_iter_ : ndarray of shape (1,) or (n_class * (n_class - 1) // 2,)
         Number of iterations run by the optimization routine to fit the model.
-        If `classes_ <= 2`, only one model is optimized, thus an integer is
-        returned. Otherwise, multiple models are optimized separately, thus
-        having multiple number of iterations. In this case a numpy array is
-        returned with the number of iterations for each model.
+        The shape of this attribute depends on the number of models optimized.
+        If `classes_ <= 2`, only one model is optimized, thus the shape of the
+        attribute is `(1,)`.
+        Otherwise, a series of models are optimized separately, having a number
+        of iterations for each. In this case the shape of the attribute is
+        `(n_class * (n_class - 1) // 2,)`.
 
         .. versionadded:: 1.1
 
@@ -934,12 +936,14 @@ class NuSVC(BaseSVC):
 
         .. versionadded:: 1.0
 
-    n_iter_ : int or ndarray of shape (n_class * (n_class - 1) // 2,)
+    n_iter_ : ndarray of shape (1,) or (n_class * (n_class - 1) // 2,)
         Number of iterations run by the optimization routine to fit the model.
-        If `classes_ <= 2`, only one model is optimized, thus an integer is
-        returned. Otherwise, multiple models are optimized separately, thus
-        having multiple number of iterations. In this case a numpy array is
-        returned with the number of iterations for each model.
+        The shape of this attribute depends on the number of models optimized.
+        If `classes_ <= 2`, only one model is optimized, thus the shape of the
+        attribute is `(1,)`.
+        Otherwise, a series of models are optimized separately, having a number
+        of iterations for each. In this case the shape of the attribute is
+        `(n_class * (n_class - 1) // 2,)`.
 
         .. versionadded:: 1.1
 

From 8e4191707dc59c12ce0531f8ed3a60c2bd2214a5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Mon, 8 Nov 2021 12:40:02 -0300
Subject: [PATCH 23/28] Apply reviewers suggestions

- Remove n_iter from set_model, predict, predict_proba, decision_function, and free_model.
- Update documentation of sub-classes SVC and NuSVC.
- Use a dictionary to load parameters in the test.
---
 sklearn/svm/_base.py                          |  6 ---
 sklearn/svm/_classes.py                       | 20 +++----
 sklearn/svm/_libsvm.pxi                       |  2 +-
 sklearn/svm/_libsvm.pyx                       | 18 ++-----
 sklearn/svm/_libsvm_sparse.pyx                | 17 +++---
 sklearn/svm/src/libsvm/libsvm_helper.c        | 19 ++++---
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c | 14 +++--
 sklearn/svm/tests/test_svm.py                 | 54 +++++++++----------
 8 files changed, 58 insertions(+), 92 deletions(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index 0195b033284ed..bcc3f23b115ef 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -460,7 +460,6 @@ def _dense_predict(self, X):
             self._n_support,
             self._dual_coef_,
             self._intercept_,
-            self._num_iter,
             self._probA,
             self._probB,
             svm_type=svm_type,
@@ -505,7 +504,6 @@ def _sparse_predict(self, X):
             self._n_support,
             self._probA,
             self._probB,
-            self._num_iter,
         )
 
     def _compute_kernel(self, X):
@@ -563,7 +561,6 @@ def _dense_decision_function(self, X):
             self._n_support,
             self._dual_coef_,
             self._intercept_,
-            self._num_iter,
             self._probA,
             self._probB,
             svm_type=LIBSVM_IMPL.index(self._impl),
@@ -607,7 +604,6 @@ def _sparse_decision_function(self, X):
             self._n_support,
             self._probA,
             self._probB,
-            self._num_iter,
         )
 
     def _validate_for_predict(self, X):
@@ -911,7 +907,6 @@ def _dense_predict_proba(self, X):
             self._n_support,
             self._dual_coef_,
             self._intercept_,
-            self._num_iter,
             self._probA,
             self._probB,
             svm_type=svm_type,
@@ -957,7 +952,6 @@ def _sparse_predict_proba(self, X):
             self._n_support,
             self._probA,
             self._probB,
-            self._num_iter,
         )
 
     def _get_coef(self):
diff --git a/sklearn/svm/_classes.py b/sklearn/svm/_classes.py
index fcd0f43984c9e..cafaf9b2c2cf5 100644
--- a/sklearn/svm/_classes.py
+++ b/sklearn/svm/_classes.py
@@ -670,14 +670,10 @@ class SVC(BaseSVC):
 
         .. versionadded:: 1.0
 
-    n_iter_ : ndarray of shape (1,) or (n_class * (n_class - 1) // 2,)
+    n_iter_ : ndarray of shape (n_classes * (n_classes - 1) // 2,)
         Number of iterations run by the optimization routine to fit the model.
-        The shape of this attribute depends on the number of models optimized.
-        If `classes_ <= 2`, only one model is optimized, thus the shape of the
-        attribute is `(1,)`.
-        Otherwise, a series of models are optimized separately, having a number
-        of iterations for each. In this case the shape of the attribute is
-        `(n_class * (n_class - 1) // 2,)`.
+        The shape of this attribute depends on the number of models optimized
+        which in turn depends on the number of classes.
 
         .. versionadded:: 1.1
 
@@ -936,14 +932,10 @@ class NuSVC(BaseSVC):
 
         .. versionadded:: 1.0
 
-    n_iter_ : ndarray of shape (1,) or (n_class * (n_class - 1) // 2,)
+    n_iter_ : ndarray of shape (n_classes * (n_classes - 1) // 2,)
         Number of iterations run by the optimization routine to fit the model.
-        The shape of this attribute depends on the number of models optimized.
-        If `classes_ <= 2`, only one model is optimized, thus the shape of the
-        attribute is `(1,)`.
-        Otherwise, a series of models are optimized separately, having a number
-        of iterations for each. In this case the shape of the attribute is
-        `(n_class * (n_class - 1) // 2,)`.
+        The shape of this attribute depends on the number of models optimized
+        which in turn depends on the number of classes.
 
         .. versionadded:: 1.1
 
diff --git a/sklearn/svm/_libsvm.pxi b/sklearn/svm/_libsvm.pxi
index 39cc40a0bc9ae..75a6fb55bcf8e 100644
--- a/sklearn/svm/_libsvm.pxi
+++ b/sklearn/svm/_libsvm.pxi
@@ -53,7 +53,7 @@ cdef extern from "libsvm_helper.c":
 
     svm_model *set_model (svm_parameter *, int, char *, np.npy_intp *,
                          char *, np.npy_intp *, np.npy_intp *, char *,
-                         char *, char *, char *, char *, char *)
+                         char *, char *, char *, char *)
 
     void copy_sv_coef   (char *, svm_model *)
     void copy_n_iter  (char *, svm_model *)
diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index bf46afa8e84be..32c0802b3a8c4 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -286,7 +286,6 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
             np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
             np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-            np.ndarray[int, ndim=1, mode='c'] n_iter,
             np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
             np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
             int svm_type=0, kernel='rbf', int degree=3,
@@ -318,9 +317,6 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     intercept : array of shape (n_class*(n_class-1)/2)
         Intercept in decision function.
 
-    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
-        Number of iterations run by the optimization routine to fit the model.
-
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates.
 
@@ -362,7 +358,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data, probA.data,
-                      probB.data, n_iter.data)
+                      probB.data)
     cdef BlasFunctions blas_functions
     blas_functions.dot = _dot[double]
     #TODO: use check_model
@@ -385,7 +381,6 @@ def predict_proba(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[int, ndim=1, mode='c'] n_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -427,9 +422,6 @@ def predict_proba(
     intercept : array of shape (n_class*(n_class-1)/2,)
         Intercept in decision function.
 
-    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
-        Number of iterations run by the optimization routine to fit the model.
-
     probA, probB : array of shape (n_class*(n_class-1)/2,)
         Probability estimates.
 
@@ -470,7 +462,7 @@ def predict_proba(
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data, n_iter.data)
+                      probA.data, probB.data)
 
     cdef np.npy_intp n_class = get_nr(model)
     cdef BlasFunctions blas_functions
@@ -494,7 +486,6 @@ def decision_function(
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=2, mode='c'] sv_coef,
     np.ndarray[np.float64_t, ndim=1, mode='c'] intercept,
-    np.ndarray[int, ndim=1, mode='c'] n_iter,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA=np.empty(0),
     np.ndarray[np.float64_t, ndim=1, mode='c'] probB=np.empty(0),
     int svm_type=0, kernel='rbf', int degree=3,
@@ -529,9 +520,6 @@ def decision_function(
     intercept : array, shape=[n_class*(n_class-1)/2]
         Intercept in decision function.
 
-    n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
-        Number of iterations run by the optimization routine to fit the model.
-
     probA, probB : array, shape=[n_class*(n_class-1)/2]
         Probability estimates.
 
@@ -576,7 +564,7 @@ def decision_function(
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
                       sv_coef.data, intercept.data, nSV.data,
-                      probA.data, probB.data, n_iter.data)
+                      probA.data, probB.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/_libsvm_sparse.pyx b/sklearn/svm/_libsvm_sparse.pyx
index b73c2cd487a4b..64fc69364b2ee 100644
--- a/sklearn/svm/_libsvm_sparse.pyx
+++ b/sklearn/svm/_libsvm_sparse.pyx
@@ -35,7 +35,7 @@ cdef extern from "libsvm_sparse_helper.c":
                             char *SV_indices, np.npy_intp *SV_intptr_dims,
                             char *SV_intptr,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB, char *n_iter)
+                            char *probA, char *probB)
     svm_parameter *set_parameter (int , int , int , double, double ,
                                   double , double , double , double,
                                   double, int, int, int, char *, char *, int,
@@ -241,8 +241,7 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                             shrinking, int probability,
                             np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
                             np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-                            np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-                            np.ndarray[int, ndim=1, mode='c'] n_iter):
+                            np.ndarray[np.float64_t, ndim=1, mode='c'] probB):
     """
     Predict values T given a model.
 
@@ -283,7 +282,7 @@ def libsvm_sparse_predict (np.ndarray[np.float64_t, ndim=1, mode='c'] T_data,
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data, n_iter.data)
+                          nSV.data, probA.data, probB.data)
     #TODO: use check_model
     dec_values = np.empty(T_indptr.shape[0]-1)
     cdef BlasFunctions blas_functions
@@ -319,8 +318,7 @@ def libsvm_sparse_predict_proba(
     double nu, double p, int shrinking, int probability,
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-    np.ndarray[int, ndim=1, mode='c'] n_iter):
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probB):
     """
     Predict values T given a model.
     """
@@ -341,7 +339,7 @@ def libsvm_sparse_predict_proba(
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data, n_iter.data)
+                          nSV.data, probA.data, probB.data)
     #TODO: use check_model
     cdef np.npy_intp n_class = get_nr(model)
     cdef int rv
@@ -381,8 +379,7 @@ def libsvm_sparse_decision_function(
     double nu, double p, int shrinking, int probability,
     np.ndarray[np.int32_t, ndim=1, mode='c'] nSV,
     np.ndarray[np.float64_t, ndim=1, mode='c'] probA,
-    np.ndarray[np.float64_t, ndim=1, mode='c'] probB,
-    np.ndarray[int, ndim=1, mode='c'] n_iter):
+    np.ndarray[np.float64_t, ndim=1, mode='c'] probB):
     """
     Predict margin (libsvm name for this is predict_values)
 
@@ -407,7 +404,7 @@ def libsvm_sparse_decision_function(
                           SV_indices.shape, SV_indices.data,
                           SV_indptr.shape, SV_indptr.data,
                           sv_coef.data, intercept.data,
-                          nSV.data, probA.data, probB.data, n_iter.data)
+                          nSV.data, probA.data, probB.data)
 
     if svm_type > 1:
         n_class = 1
diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index f9796f81e4b72..2f6b0c99ef98b 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -116,14 +116,13 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
                             char *support, npy_intp *support_dims,
                             npy_intp *sv_coef_strides,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB, char *n_iter)
+                            char *probA, char *probB)
 {
     struct svm_model *model;
     double *dsv_coef = (double *) sv_coef;
-    int i, m, n_models;
+    int i, m;
 
     m = nr_class * (nr_class-1)/2;
-    n_models = nr_class <= 2 ? 1 : m;
 
     if ((model = malloc(sizeof(struct svm_model))) == NULL)
         goto model_error;
@@ -135,8 +134,9 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
         goto sv_coef_error;
     if ((model->rho = malloc( m * sizeof(double))) == NULL)
         goto rho_error;
-    if ((model->n_iter = malloc(n_models * sizeof(int))) == NULL)
-        goto n_iter_error;
+
+    // This is only allocated in dynamic memory while training.
+    model->n_iter = NULL;
 
     model->nr_class = nr_class;
     model->param = *param;
@@ -191,8 +191,6 @@ struct svm_model *set_model(struct svm_parameter *param, int nr_class,
     model->free_sv = 0;
     return model;
 
-n_iter_error:
-    free(model->n_iter);
 probB_error:
     free(model->probA);
 probA_error:
@@ -384,15 +382,16 @@ int free_model(struct svm_model *model)
     if (model == NULL) return -1;
     free(model->SV);
 
-    /* We don't free sv_ind, since we did not create them in
+    /* We don't free sv_ind and n_iter, since we did not create them in
        set_model */
-    /* free(model->sv_ind); */
+    /* free(model->sv_ind);
+     * free(model->n_iter);
+     */
     free(model->sv_coef);
     free(model->rho);
     free(model->label);
     free(model->probA);
     free(model->probB);
-    free(model->n_iter);
     free(model->nSV);
     free(model);
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index 5ad2106550027..069365b47204b 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -109,14 +109,13 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
                             char *SV_indices, npy_intp *SV_indptr_dims,
                             char *SV_intptr,
                             char *sv_coef, char *rho, char *nSV,
-                            char *probA, char *probB, char *n_iter)
+                            char *probA, char *probB)
 {
     struct svm_csr_model *model;
     double *dsv_coef = (double *) sv_coef;
-    int i, m, n_models;
+    int i, m;
 
     m = nr_class * (nr_class-1)/2;
-    n_models = nr_class <= 2 ? 1 : m;
 
     if ((model = malloc(sizeof(struct svm_csr_model))) == NULL)
         goto model_error;
@@ -128,8 +127,9 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
         goto sv_coef_error;
     if ((model->rho = malloc( m * sizeof(double))) == NULL)
         goto rho_error;
-    if ((model->n_iter = malloc(n_models * sizeof(int))) == NULL)
-        goto n_iter_error;
+
+    // This is only allocated in dynamic memory while training.
+    model->n_iter = NULL;
 
     /* in the case of precomputed kernels we do not use
        dense_to_precomputed because we don't want the leading 0. As
@@ -189,8 +189,6 @@ struct svm_csr_model *csr_set_model(struct svm_parameter *param, int nr_class,
     model->free_sv = 0;
     return model;
 
-n_iter_error:
-    free(model->n_iter);
 probB_error:
     free(model->probA);
 probA_error:
@@ -422,6 +420,7 @@ int free_problem(struct svm_csr_problem *problem)
 int free_model(struct svm_csr_model *model)
 {
     /* like svm_free_and_destroy_model, but does not free sv_coef[i] */
+    /* We don't free n_iter, since we did not create them in set_model. */
     if (model == NULL) return -1;
     free(model->SV);
     free(model->sv_coef);
@@ -429,7 +428,6 @@ int free_model(struct svm_csr_model *model)
     free(model->label);
     free(model->probA);
     free(model->probB);
-    free(model->n_iter);
     free(model->nSV);
     free(model);
 
diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index 5bc6a1e92b736..c52f4fddc6158 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -66,8 +66,8 @@ def test_libsvm_iris():
     assert_array_equal(clf.classes_, np.sort(clf.classes_))
 
     # check also the low-level API
-    # We unpack the values to be able to repack some of the
-    # return values from libsvm.
+    # We unpack the values to create a dictionary with some of the return values
+    # from Libsvm's fit.
     (
         libsvm_support,
         libsvm_support_vectors,
@@ -76,26 +76,25 @@ def test_libsvm_iris():
         libsvm_intercept,
         libsvm_probA,
         libsvm_probB,
-        # libsvm_fit_status won't be packed bellow.
+        # libsvm_fit_status and libsvm_n_iter won't be used below.
         libsvm_fit_status,
         libsvm_n_iter,
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
 
-    lib_svm_model = (
-        libsvm_support,
-        libsvm_support_vectors,
-        libsvm_n_class_SV,
-        libsvm_sv_coef,
-        libsvm_intercept,
-        libsvm_n_iter,
-        libsvm_probA,
-        libsvm_probB,
-    )
-    pred = _libsvm.predict(iris.data, *lib_svm_model)
+    model_params = {
+        'support': libsvm_support,
+        'SV': libsvm_support_vectors,
+        'nSV': libsvm_n_class_SV,
+        'sv_coef': libsvm_sv_coef,
+        'intercept': libsvm_intercept,
+        'probA': libsvm_probA,
+        'probB': libsvm_probB,
+    }
+    pred = _libsvm.predict(iris.data, **model_params)
     assert np.mean(pred == iris.target) > 0.95
 
-    # We unpack the values to be able to repack some of the
-    # return values from libsvm.
+    # We unpack the values to create a dictionary with some of the return values
+    # from Libsvm's fit.
     (
         libsvm_support,
         libsvm_support_vectors,
@@ -104,22 +103,21 @@ def test_libsvm_iris():
         libsvm_intercept,
         libsvm_probA,
         libsvm_probB,
-        # libsvm_fit_status won't be packed bellow.
+        # libsvm_fit_status and libsvm_n_iter won't be used below.
         libsvm_fit_status,
         libsvm_n_iter,
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
 
-    lib_svm_model = (
-        libsvm_support,
-        libsvm_support_vectors,
-        libsvm_n_class_SV,
-        libsvm_sv_coef,
-        libsvm_intercept,
-        libsvm_n_iter,
-        libsvm_probA,
-        libsvm_probB,
-    )
-    pred = _libsvm.predict(iris.data, *lib_svm_model, kernel="linear")
+    model_params = {
+        'support': libsvm_support,
+        'SV': libsvm_support_vectors,
+        'nSV': libsvm_n_class_SV,
+        'sv_coef': libsvm_sv_coef,
+        'intercept': libsvm_intercept,
+        'probA': libsvm_probA,
+        'probB': libsvm_probB,
+    }
+    pred = _libsvm.predict(iris.data, **model_params, kernel="linear")
     assert np.mean(pred == iris.target) > 0.95
 
     pred = _libsvm.cross_validation(

From b82e355f3f2d62728ae84d60c7f57dcb3e36bb36 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Mon, 8 Nov 2021 12:59:16 -0300
Subject: [PATCH 24/28] Apply black to sklearn/svm/tests/test_svm.py

---
 sklearn/svm/tests/test_svm.py | 28 ++++++++++++++--------------
 1 file changed, 14 insertions(+), 14 deletions(-)

diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index 24dcde8db9b20..90d7a39d0e340 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -82,13 +82,13 @@ def test_libsvm_iris():
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64))
 
     model_params = {
-        'support': libsvm_support,
-        'SV': libsvm_support_vectors,
-        'nSV': libsvm_n_class_SV,
-        'sv_coef': libsvm_sv_coef,
-        'intercept': libsvm_intercept,
-        'probA': libsvm_probA,
-        'probB': libsvm_probB,
+        "support": libsvm_support,
+        "SV": libsvm_support_vectors,
+        "nSV": libsvm_n_class_SV,
+        "sv_coef": libsvm_sv_coef,
+        "intercept": libsvm_intercept,
+        "probA": libsvm_probA,
+        "probB": libsvm_probB,
     }
     pred = _libsvm.predict(iris.data, **model_params)
     assert np.mean(pred == iris.target) > 0.95
@@ -109,13 +109,13 @@ def test_libsvm_iris():
     ) = _libsvm.fit(iris.data, iris.target.astype(np.float64), kernel="linear")
 
     model_params = {
-        'support': libsvm_support,
-        'SV': libsvm_support_vectors,
-        'nSV': libsvm_n_class_SV,
-        'sv_coef': libsvm_sv_coef,
-        'intercept': libsvm_intercept,
-        'probA': libsvm_probA,
-        'probB': libsvm_probB,
+        "support": libsvm_support,
+        "SV": libsvm_support_vectors,
+        "nSV": libsvm_n_class_SV,
+        "sv_coef": libsvm_sv_coef,
+        "intercept": libsvm_intercept,
+        "probA": libsvm_probA,
+        "probB": libsvm_probB,
     }
     pred = _libsvm.predict(iris.data, **model_params, kernel="linear")
     assert np.mean(pred == iris.target) > 0.95

From 31df365e9d52da922fea6edf90f54c0dfa887de4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Wed, 10 Nov 2021 21:10:32 -0300
Subject: [PATCH 25/28] Apply reviewers suggestions

Co-authored-by: Adrin Jalali <adrin.jalali@gmail.com>
---
 sklearn/svm/src/libsvm/libsvm_helper.c        | 2 +-
 sklearn/svm/src/libsvm/libsvm_sparse_helper.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/sklearn/svm/src/libsvm/libsvm_helper.c b/sklearn/svm/src/libsvm/libsvm_helper.c
index 2f6b0c99ef98b..1adf6b1b35370 100644
--- a/sklearn/svm/src/libsvm/libsvm_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_helper.c
@@ -233,7 +233,7 @@ npy_intp get_nr(struct svm_model *model)
  */
 void copy_n_iter(char *data, struct svm_model *model)
 {
-    const int n_models = MAX(1, model->nr_class * (model->nr_class-1)/2);
+    const int n_models = MAX(1, model->nr_class * (model->nr_class-1) / 2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 
diff --git a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
index 069365b47204b..08556212bab5e 100644
--- a/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
+++ b/sklearn/svm/src/libsvm/libsvm_sparse_helper.c
@@ -362,7 +362,7 @@ void copy_sv_coef(char *data, struct svm_csr_model *model)
  */
 void copy_n_iter(char *data, struct svm_csr_model *model)
 {
-    const int n_models = MAX(1, model->nr_class * (model->nr_class-1)/2);
+    const int n_models = MAX(1, model->nr_class * (model->nr_class-1) / 2);
     memcpy(data, model->n_iter, n_models * sizeof(int));
 }
 

From babbb2c5711d5a3d23404c03c7ba4fbe6bced1db Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Wed, 10 Nov 2021 22:51:17 -0300
Subject: [PATCH 26/28] Revert formatting for unchanged line

---
 sklearn/svm/_libsvm.pyx | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/sklearn/svm/_libsvm.pyx b/sklearn/svm/_libsvm.pyx
index 32c0802b3a8c4..4df99724b790a 100644
--- a/sklearn/svm/_libsvm.pyx
+++ b/sklearn/svm/_libsvm.pyx
@@ -357,8 +357,7 @@ def predict(np.ndarray[np.float64_t, ndim=2, mode='c'] X,
                        class_weight_label.data, class_weight.data)
     model = set_model(&param, <int> nSV.shape[0], SV.data, SV.shape,
                       support.data, support.shape, sv_coef.strides,
-                      sv_coef.data, intercept.data, nSV.data, probA.data,
-                      probB.data)
+                      sv_coef.data, intercept.data, nSV.data, probA.data, probB.data)
     cdef BlasFunctions blas_functions
     blas_functions.dot = _dot[double]
     #TODO: use check_model

From d4fa41d02d6d15c6953f243c83a6290159726b26 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Sat, 11 Dec 2021 17:20:36 -0300
Subject: [PATCH 27/28] Convert n_iter_ to int

---
 sklearn/svm/_base.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sklearn/svm/_base.py b/sklearn/svm/_base.py
index bcc3f23b115ef..2c74ae153543b 100644
--- a/sklearn/svm/_base.py
+++ b/sklearn/svm/_base.py
@@ -283,7 +283,7 @@ def fit(self, X, y, sample_weight=None):
         if self._impl in ["c_svc", "nu_svc"]:
             self.n_iter_ = self._num_iter
         else:
-            self.n_iter_ = self._num_iter[0]
+            self.n_iter_ = self._num_iter.item()
 
         return self
 

From ff3fabded6062af33199f5486e7baf7869b6d322 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Mart=C3=ADn=20Loyola?= <jmloyola@outlook.com>
Date: Mon, 13 Dec 2021 19:40:00 -0300
Subject: [PATCH 28/28] Add coverage tests

---
 sklearn/svm/tests/test_svm.py | 39 ++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)

diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
index 90d7a39d0e340..af4e7d4a0935b 100644
--- a/sklearn/svm/tests/test_svm.py
+++ b/sklearn/svm/tests/test_svm.py
@@ -1105,16 +1105,17 @@ def test_svc_bad_kernel():
         svc.fit(X, Y)
 
 
-def test_timeout():
+def test_libsvm_convergence_warnings():
     a = svm.SVC(
-        kernel=lambda x, y: np.dot(x, y.T), probability=True, random_state=0, max_iter=1
+        kernel=lambda x, y: np.dot(x, y.T), probability=True, random_state=0, max_iter=2
     )
     warning_msg = (
-        r"Solver terminated early \(max_iter=1\).  Consider pre-processing "
+        r"Solver terminated early \(max_iter=2\).  Consider pre-processing "
         r"your data with StandardScaler or MinMaxScaler."
     )
     with pytest.warns(ConvergenceWarning, match=warning_msg):
         a.fit(np.array(X), Y)
+    assert np.all(a.n_iter_ == 2)
 
 
 def test_unfitted():
@@ -1468,3 +1469,35 @@ def test_svc_raises_error_internal_representation():
     msg = "The internal representation of SVC was altered"
     with pytest.raises(ValueError, match=msg):
         clf.predict(X)
+
+
+@pytest.mark.parametrize(
+    "estimator, expected_n_iter_type",
+    [
+        (svm.SVC, np.ndarray),
+        (svm.NuSVC, np.ndarray),
+        (svm.SVR, int),
+        (svm.NuSVR, int),
+        (svm.OneClassSVM, int),
+    ],
+)
+@pytest.mark.parametrize(
+    "dataset",
+    [
+        make_classification(n_classes=2, n_informative=2, random_state=0),
+        make_classification(n_classes=3, n_informative=3, random_state=0),
+        make_classification(n_classes=4, n_informative=4, random_state=0),
+    ],
+)
+def test_n_iter_libsvm(estimator, expected_n_iter_type, dataset):
+    # Check that the type of n_iter_ is correct for the classes that inherit
+    # from BaseSVC.
+    # Note that for SVC, and NuSVC this is an ndarray; while for SVR, NuSVR, and
+    # OneClassSVM, it is an int.
+    # For SVC and NuSVC also check the shape of n_iter_.
+    X, y = dataset
+    n_iter = estimator(kernel="linear").fit(X, y).n_iter_
+    assert type(n_iter) == expected_n_iter_type
+    if estimator in [svm.SVC, svm.NuSVC]:
+        n_classes = len(np.unique(y))
+        assert n_iter.shape == (n_classes * (n_classes - 1) // 2,)