Skip to content

ENH Add dtype preservation for SpectralClustering #22669

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 7 commits into from
Closed

ENH Add dtype preservation for SpectralClustering #22669

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
582aa0b
TST Adapt test_spectral.py to test implementations on 32bit datasets
jjerphan Mar 3, 2022
2a0b747
Merge branch 'main' into tst/test_spectral-32bit
jjerphan Mar 23, 2022
818e85c
TST Use global_dtype
jjerphan Mar 23, 2022
dcc03f5
Address reviews comments
jjerphan Mar 29, 2022
b60a65f
Merge branch 'main' into tst/test_spectral-32bit
jjerphan Nov 16, 2022
1d05d2d
Make sure to preserve dtype in SpectralClustering
jjerphan Nov 16, 2022
ac6dce4
Merge branch 'main' into tst/test_spectral-32bit
jjerphan Nov 23, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions sklearn/cluster/_spectral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -744,7 +744,7 @@ def fit(self, X, y=None):
params["coef0"] = self.coef0
self.affinity_matrix_ = pairwise_kernels(
X, metric=self.affinity, filter_params=True, **params
)
).astype(X.dtype, copy=False)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we should work on making pairwise_kernels work efficiently with float32 data first.

Since affinity="rbf" is the default, flagging SpectralClustering as dtype preserving without that prerequisite would be misleading to our users: there would be very few performance or peak memory usage gains in passing float32 data to this estimator when using the default hyper-params.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

BTW we also need a test that checks the dtype of affinity_matrix_ when the affinity param is the str name of a kernel function and another such assertion in a test that covers the case when affinity is nearest_neighbors.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We also need to have ArgKmin and RadiusNeighbors preserve dtypes because it is also used when affinity="precomputed_nearest_neighbors".

Let's turn this PR as a draft for now.


random_state = check_random_state(self.random_state)
self.labels_ = spectral_clustering(
Expand Down Expand Up @@ -787,5 +787,6 @@ def fit_predict(self, X, y=None):
def _more_tags(self):
return {
"pairwise": self.affinity
in ["precomputed", "precomputed_nearest_neighbors"]
in ["precomputed", "precomputed_nearest_neighbors"],
"preserves_dtype": [np.float64, np.float32],
}
25 changes: 15 additions & 10 deletions sklearn/cluster/tests/test_spectral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@

@pytest.mark.parametrize("eigen_solver", ("arpack", "lobpcg"))
@pytest.mark.parametrize("assign_labels", ("kmeans", "discretize", "cluster_qr"))
def test_spectral_clustering(eigen_solver, assign_labels):
def test_spectral_clustering(eigen_solver, assign_labels, global_dtype):
S = np.array(
[
[1.0, 1.0, 1.0, 0.2, 0.0, 0.0, 0.0],
Expand All @@ -50,7 +50,8 @@ def test_spectral_clustering(eigen_solver, assign_labels):
[0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0],
[0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0],
[0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0],
]
],
dtype=global_dtype,
)

for mat in (S, sparse.csr_matrix(S)):
Expand All @@ -74,14 +75,14 @@ def test_spectral_clustering(eigen_solver, assign_labels):


@pytest.mark.parametrize("assign_labels", ("kmeans", "discretize", "cluster_qr"))
def test_spectral_clustering_sparse(assign_labels):
def test_spectral_clustering_sparse(assign_labels, global_dtype):
X, y = make_blobs(
n_samples=20, random_state=0, centers=[[1, 1], [-1, -1]], cluster_std=0.01
)

S = rbf_kernel(X, gamma=1)
S = np.maximum(S - 1e-4, 0)
S = sparse.coo_matrix(S)
S = sparse.coo_matrix(S, dtype=global_dtype)

labels = (
SpectralClustering(
Expand All @@ -96,11 +97,12 @@ def test_spectral_clustering_sparse(assign_labels):
assert adjusted_rand_score(y, labels) == 1
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please check the dtype of affinity_matrix_ here.



def test_precomputed_nearest_neighbors_filtering():
def test_precomputed_nearest_neighbors_filtering(global_dtype):
# Test precomputed graph filtering when containing too many neighbors
X, y = make_blobs(
n_samples=200, random_state=0, centers=[[1, 1], [-1, -1]], cluster_std=0.01
)
X = X.astype(global_dtype, copy=False)

n_neighbors = 2
results = []
Expand All @@ -122,13 +124,14 @@ def test_precomputed_nearest_neighbors_filtering():
assert_array_equal(results[0], results[1])
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please check the dtype of affinity_matrix_ here.



def test_affinities():
def test_affinities(global_dtype):
# Note: in the following, random_state has been selected to have
# a dataset that yields a stable eigen decomposition both when built
# on OSX and Linux
X, y = make_blobs(
n_samples=20, random_state=0, centers=[[1, 1], [-1, -1]], cluster_std=0.01
)
X = X.astype(global_dtype, copy=False)
# nearest neighbors affinity
sp = SpectralClustering(n_clusters=2, affinity="nearest_neighbors", random_state=0)
with pytest.warns(UserWarning, match="not fully connected"):
Expand All @@ -140,6 +143,7 @@ def test_affinities():
assert adjusted_rand_score(y, labels) == 1

X = check_random_state(10).rand(10, 5) * 10
X = X.astype(global_dtype, copy=False)

kernels_available = kernel_metrics()
for kern in kernels_available:
Expand Down Expand Up @@ -182,13 +186,13 @@ def test_cluster_qr():
assert np.array_equal(labels_float64, labels_float32)


def test_cluster_qr_permutation_invariance():
def test_cluster_qr_permutation_invariance(global_dtype):
# cluster_qr must be invariant to sample permutation.
random_state = np.random.RandomState(seed=8)
n_samples, n_components = 100, 5
data = random_state.randn(n_samples, n_components)
data = random_state.randn(n_samples, n_components).astype(global_dtype, copy=False)
perm = random_state.permutation(n_samples)
assert np.array_equal(
assert_array_equal(
cluster_qr(data)[perm],
cluster_qr(data[perm]),
)
Expand Down Expand Up @@ -263,12 +267,13 @@ def test_spectral_clustering_with_arpack_amg_solvers():
spectral_clustering(graph, n_clusters=2, eigen_solver="amg", random_state=0)


def test_n_components():
def test_n_components(global_dtype):
# Test that after adding n_components, result is different and
# n_components = n_clusters by default
X, y = make_blobs(
n_samples=20, random_state=0, centers=[[1, 1], [-1, -1]], cluster_std=0.01
)
X = X.astype(global_dtype, copy=False)
sp = SpectralClustering(n_clusters=2, random_state=0)
labels = sp.fit(X).labels_
# set n_components = n_cluster and test if result is the same
Expand Down