scikit-learn-contrib · bellet · Jan 29, 2019 · Jan 23, 2019 · Jan 24, 2019 · Jan 29, 2019
diff --git a/metric_learn/itml.py b/metric_learn/itml.py
@@ -73,9 +73,9 @@ def _fit(self, pairs, y, bounds=None):
     self.bounds_[self.bounds_==0] = 1e-9
     # init metric
     if self.A0 is None:
-      self.A_ = np.identity(pairs.shape[2])
+      A = np.identity(pairs.shape[2])
     else:
-      self.A_ = check_array(self.A0)
+      A = check_array(self.A0, copy=True)
     gamma = self.gamma
     pos_pairs, neg_pairs = pairs[y == 1], pairs[y == -1]
     num_pos = len(pos_pairs)
@@ -87,7 +87,6 @@ def _fit(self, pairs, y, bounds=None):
     neg_bhat = np.zeros(num_neg) + self.bounds_[1]
     pos_vv = pos_pairs[:, 0, :] - pos_pairs[:, 1, :]
     neg_vv = neg_pairs[:, 0, :] - neg_pairs[:, 1, :]
-    A = self.A_
 
     for it in xrange(self.max_iter):
       # update positives
@@ -125,7 +124,7 @@ def _fit(self, pairs, y, bounds=None):
       print('itml converged at iter: %d, conv = %f' % (it, conv))
     self.n_iter_ = it
 
-    self.transformer_ = transformer_from_metric(self.A_)
+    self.transformer_ = transformer_from_metric(A)
     return self
 
 
@@ -134,6 +133,18 @@ class ITML(_BaseITML, _PairsClassifierMixin):
 
   Attributes
   ----------
+  bounds_ : array-like, shape=(2,)
+      Bounds on similarity, aside slack variables, s.t.
+      ``d(a, b) < bounds_[0]`` for all given pairs of similar points ``a``
+      and ``b``, and ``d(c, d) > bounds_[1]`` for all given pairs of
+      dissimilar points ``c`` and ``d``, with ``d`` the learned distance. If
+      not provided at initialization, bounds_[0] and bounds_[1] are set at
+      train time to the 5th and 95th percentile of the pairwise distances among
+      all points present in the input `pairs`.
+
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The linear transformation ``L`` deduced from the learned Mahalanobis
       metric (See function `transformer_from_metric`.)
@@ -151,8 +162,14 @@ def fit(self, pairs, y, bounds=None):
         preprocessor.
     y: array-like, of shape (n_constraints,)
         Labels of constraints. Should be -1 for dissimilar pair, 1 for similar.
-    bounds : list (pos,neg) pairs, optional
-        bounds on similarity, s.t. d(X[a],X[b]) < pos and d(X[c],X[d]) > neg
+    bounds : `list` of two numbers
+        Bounds on similarity, aside slack variables, s.t.
+        ``d(a, b) < bounds_[0]`` for all given pairs of similar points ``a``
+        and ``b``, and ``d(c, d) > bounds_[1]`` for all given pairs of
+        dissimilar points ``c`` and ``d``, with ``d`` the learned distance.
+        If not provided at initialization, bounds_[0] and bounds_[1] will be
+        set to the 5th and 95th percentile of the pairwise distances among all
+        points present in the input `pairs`.
 
     Returns
     -------
@@ -167,6 +184,18 @@ class ITML_Supervised(_BaseITML, TransformerMixin):
 
   Attributes
   ----------
+  bounds_ : array-like, shape=(2,)
+      Bounds on similarity, aside slack variables, s.t.
+      ``d(a, b) < bounds_[0]`` for all given pairs of similar points ``a``
+      and ``b``, and ``d(c, d) > bounds_[1]`` for all given pairs of
+      dissimilar points ``c`` and ``d``, with ``d`` the learned distance.
+      If not provided at initialization, bounds_[0] and bounds_[1] are set at
+      train time to the 5th and 95th percentile of the pairwise distances
+      among all points in the training data `X`.
+
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The linear transformation ``L`` deduced from the learned Mahalanobis
       metric (See function `transformer_from_metric`.)
@@ -193,8 +222,14 @@ def __init__(self, gamma=1., max_iter=1000, convergence_threshold=1e-3,
              be removed in 0.6.0.
     num_constraints: int, optional
         number of constraints to generate
-    bounds : list (pos,neg) pairs, optional
-        bounds on similarity, s.t. d(X[a],X[b]) < pos and d(X[c],X[d]) > neg
+    bounds : `list` of two numbers
+      Bounds on similarity, aside slack variables, s.t.
+      ``d(a, b) < bounds_[0]`` for all given pairs of similar points ``a``
+      and ``b``, and ``d(c, d) > bounds_[1]`` for all given pairs of
+      dissimilar points ``c`` and ``d``, with ``d`` the learned distance.
+      If not provided at initialization, bounds_[0] and bounds_[1] will be
+      set to the 5th and 95th percentile of the pairwise distances among all
+      points in the training data `X`.
     A0 : (d x d) matrix, optional
         initial regularization matrix, defaults to identity
     verbose : bool, optional

diff --git a/metric_learn/lmnn.py b/metric_learn/lmnn.py
@@ -60,20 +60,20 @@ def fit(self, X, y):
     X, y = self._prepare_inputs(X, y, dtype=float,
                                 ensure_min_samples=2)
     num_pts, num_dims = X.shape
-    unique_labels, self.label_inds_ = np.unique(y, return_inverse=True)
-    if len(self.label_inds_) != num_pts:
+    unique_labels, label_inds = np.unique(y, return_inverse=True)
+    if len(label_inds) != num_pts:
       raise ValueError('Must have one label per point.')
     self.labels_ = np.arange(len(unique_labels))
     if self.use_pca:
       warnings.warn('use_pca does nothing for the python_LMNN implementation')
     self.transformer_ = np.eye(num_dims)
-    required_k = np.bincount(self.label_inds_).min()
+    required_k = np.bincount(label_inds).min()
     if self.k > required_k:
       raise ValueError('not enough class labels for specified k'
                        ' (smallest class has %d)' % required_k)
 
-    target_neighbors = self._select_targets(X)
-    impostors = self._find_impostors(target_neighbors[:, -1], X)
+    target_neighbors = self._select_targets(X, label_inds)
+    impostors = self._find_impostors(target_neighbors[:, -1], X, label_inds)
     if len(impostors) == 0:
         # L has already been initialized to an identity matrix
         return
@@ -196,23 +196,23 @@ def _loss_grad(self, X, L, dfG, impostors, it, k, reg, target_neighbors, df,
     objective += G.flatten().dot(L.T.dot(L).flatten())
     return G, objective, total_active, df, a1, a2
 
-  def _select_targets(self, X):
+  def _select_targets(self, X, label_inds):
     target_neighbors = np.empty((X.shape[0], self.k), dtype=int)
     for label in self.labels_:
-      inds, = np.nonzero(self.label_inds_ == label)
+      inds, = np.nonzero(label_inds == label)
       dd = euclidean_distances(X[inds], squared=True)
       np.fill_diagonal(dd, np.inf)
       nn = np.argsort(dd)[..., :self.k]
       target_neighbors[inds] = inds[nn]
     return target_neighbors
 
-  def _find_impostors(self, furthest_neighbors, X):
+  def _find_impostors(self, furthest_neighbors, X, label_inds):
     Lx = self.transform(X)
     margin_radii = 1 + _inplace_paired_L2(Lx[furthest_neighbors], Lx)
     impostors = []
     for label in self.labels_[:-1]:
-      in_inds, = np.nonzero(self.label_inds_ == label)
-      out_inds, = np.nonzero(self.label_inds_ > label)
+      in_inds, = np.nonzero(label_inds == label)
+      out_inds, = np.nonzero(label_inds > label)
       dist = euclidean_distances(Lx[out_inds], Lx[in_inds], squared=True)
       i1,j1 = np.nonzero(dist < margin_radii[out_inds][:,None])
       i2,j2 = np.nonzero(dist < margin_radii[in_inds])
@@ -265,6 +265,9 @@ class LMNN(_base_LMNN):
 
     Attributes
     ----------
+    n_iter_ : `int`
+        The number of iterations the solver has run.
+
     transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
         The learned linear transformation ``L``.
     """

diff --git a/metric_learn/lsml.py b/metric_learn/lsml.py
@@ -50,32 +50,32 @@ def _fit(self, quadruplets, y=None, weights=None):
                                        type_of_inputs='tuples')
 
     # check to make sure that no two constrained vectors are identical
-    self.vab_ = quadruplets[:, 0, :] - quadruplets[:, 1, :]
-    self.vcd_ = quadruplets[:, 2, :] - quadruplets[:, 3, :]
-    if self.vab_.shape != self.vcd_.shape:
+    vab = quadruplets[:, 0, :] - quadruplets[:, 1, :]
+    vcd = quadruplets[:, 2, :] - quadruplets[:, 3, :]
+    if vab.shape != vcd.shape:
       raise ValueError('Constraints must have same length')
     if weights is None:
-      self.w_ = np.ones(self.vab_.shape[0])
+      self.w_ = np.ones(vab.shape[0])
     else:
       self.w_ = weights
     self.w_ /= self.w_.sum()  # weights must sum to 1
     if self.prior is None:
       X = np.vstack({tuple(row) for row in
                      quadruplets.reshape(-1, quadruplets.shape[2])})
-      self.prior_inv_ = np.atleast_2d(np.cov(X, rowvar=False))
-      self.M_ = np.linalg.inv(self.prior_inv_)
+      prior_inv = np.atleast_2d(np.cov(X, rowvar=False))
+      M = np.linalg.inv(prior_inv)
     else:
-      self.M_ = self.prior
-      self.prior_inv_ = np.linalg.inv(self.prior)
+      M = self.prior
+      prior_inv = np.linalg.inv(self.prior)
 
     step_sizes = np.logspace(-10, 0, 10)
     # Keep track of the best step size and the loss at that step.
     l_best = 0
-    s_best = self._total_loss(self.M_)
+    s_best = self._total_loss(M, vab, vcd, prior_inv)
     if self.verbose:
       print('initial loss', s_best)
     for it in xrange(1, self.max_iter+1):
-      grad = self._gradient(self.M_)
+      grad = self._gradient(M, vab, vcd, prior_inv)
       grad_norm = scipy.linalg.norm(grad)
       if grad_norm < self.tol:
         break
@@ -84,10 +84,10 @@ def _fit(self, quadruplets, y=None, weights=None):
       M_best = None
       for step_size in step_sizes:
         step_size /= grad_norm
-        new_metric = self.M_ - step_size * grad
+        new_metric = M - step_size * grad
         w, v = scipy.linalg.eigh(new_metric)
         new_metric = v.dot((np.maximum(w, 1e-8) * v).T)
-        cur_s = self._total_loss(new_metric)
+        cur_s = self._total_loss(new_metric, vab, vcd, prior_inv)
         if cur_s < s_best:
           l_best = step_size
           s_best = cur_s
@@ -96,36 +96,36 @@ def _fit(self, quadruplets, y=None, weights=None):
         print('iter', it, 'cost', s_best, 'best step', l_best * grad_norm)
       if M_best is None:
         break
-      self.M_ = M_best
+      M = M_best
     else:
       if self.verbose:
         print("Didn't converge after", it, "iterations. Final loss:", s_best)
     self.n_iter_ = it
 
-    self.transformer_ = transformer_from_metric(self.M_)
+    self.transformer_ = transformer_from_metric(M)
     return self
 
-  def _comparison_loss(self, metric):
-    dab = np.sum(self.vab_.dot(metric) * self.vab_, axis=1)
-    dcd = np.sum(self.vcd_.dot(metric) * self.vcd_, axis=1)
+  def _comparison_loss(self, metric, vab, vcd):
+    dab = np.sum(vab.dot(metric) * vab, axis=1)
+    dcd = np.sum(vcd.dot(metric) * vcd, axis=1)
     violations = dab > dcd
     return self.w_[violations].dot((np.sqrt(dab[violations]) -
                                     np.sqrt(dcd[violations]))**2)
 
-  def _total_loss(self, metric):
+  def _total_loss(self, metric, vab, vcd, prior_inv):
     # Regularization loss
     sign, logdet = np.linalg.slogdet(metric)
-    reg_loss = np.sum(metric * self.prior_inv_) - sign * logdet
-    return self._comparison_loss(metric) + reg_loss
+    reg_loss = np.sum(metric * prior_inv) - sign * logdet
+    return self._comparison_loss(metric, vab, vcd) + reg_loss
 
-  def _gradient(self, metric):
-    dMetric = self.prior_inv_ - np.linalg.inv(metric)
-    dabs = np.sum(self.vab_.dot(metric) * self.vab_, axis=1)
-    dcds = np.sum(self.vcd_.dot(metric) * self.vcd_, axis=1)
+  def _gradient(self, metric, vab, vcd, prior_inv):
+    dMetric = prior_inv - np.linalg.inv(metric)
+    dabs = np.sum(vab.dot(metric) * vab, axis=1)
+    dcds = np.sum(vcd.dot(metric) * vcd, axis=1)
     violations = dabs > dcds
     # TODO: vectorize
-    for vab, dab, vcd, dcd in zip(self.vab_[violations], dabs[violations],
-                                  self.vcd_[violations], dcds[violations]):
+    for vab, dab, vcd, dcd in zip(vab[violations], dabs[violations],
+                                  vcd[violations], dcds[violations]):
       dMetric += ((1-np.sqrt(dcd/dab))*np.outer(vab, vab) +
                   (1-np.sqrt(dab/dcd))*np.outer(vcd, vcd))
     return dMetric
@@ -136,6 +136,9 @@ class LSML(_BaseLSML, _QuadrupletsClassifierMixin):
 
   Attributes
   ----------
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The linear transformation ``L`` deduced from the learned Mahalanobis
       metric (See function `transformer_from_metric`.)
@@ -169,6 +172,9 @@ class LSML_Supervised(_BaseLSML, TransformerMixin):
 
   Attributes
   ----------
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The linear transformation ``L`` deduced from the learned Mahalanobis
       metric (See function `transformer_from_metric`.)

diff --git a/metric_learn/mlkr.py b/metric_learn/mlkr.py
@@ -30,6 +30,9 @@ class MLKR(MahalanobisMixin, TransformerMixin):
 
   Attributes
   ----------
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The learned linear transformation ``L``.
   """

diff --git a/metric_learn/mmc.py b/metric_learn/mmc.py
@@ -353,6 +353,9 @@ class MMC(_BaseMMC, _PairsClassifierMixin):
 
   Attributes
   ----------
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The linear transformation ``L`` deduced from the learned Mahalanobis
       metric (See function `transformer_from_metric`.)
@@ -384,6 +387,9 @@ class MMC_Supervised(_BaseMMC, TransformerMixin):
 
   Attributes
   ----------
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The linear transformation ``L`` deduced from the learned Mahalanobis
       metric (See function `transformer_from_metric`.)

diff --git a/metric_learn/nca.py b/metric_learn/nca.py
@@ -24,6 +24,9 @@ class NCA(MahalanobisMixin, TransformerMixin):
 
   Attributes
   ----------
+  n_iter_ : `int`
+      The number of iterations the solver has run.
+
   transformer_ : `numpy.ndarray`, shape=(num_dims, n_features)
       The learned linear transformation ``L``.
   """

diff --git a/metric_learn/sdml.py b/metric_learn/sdml.py
@@ -58,18 +58,18 @@ def _fit(self, pairs, y):
     # set up prior M
     if self.use_cov:
       X = np.vstack({tuple(row) for row in pairs.reshape(-1, pairs.shape[2])})
-      self.M_ = pinvh(np.atleast_2d(np.cov(X, rowvar = False)))
+      M = pinvh(np.atleast_2d(np.cov(X, rowvar = False)))
     else:
-      self.M_ = np.identity(pairs.shape[2])
+      M = np.identity(pairs.shape[2])
     diff = pairs[:, 0] - pairs[:, 1]
     loss_matrix = (diff.T * y).dot(diff)
-    P = self.M_ + self.balance_param * loss_matrix
+    P = M + self.balance_param * loss_matrix
     emp_cov = pinvh(P)
     # hack: ensure positive semidefinite
     emp_cov = emp_cov.T.dot(emp_cov)
-    _, self.M_ = graph_lasso(emp_cov, self.sparsity_param, verbose=self.verbose)
+    _, M = graph_lasso(emp_cov, self.sparsity_param, verbose=self.verbose)
 
-    self.transformer_ = transformer_from_metric(self.M_)
+    self.transformer_ = transformer_from_metric(M)
     return self