8000 ENH Add gamma='scale' option to RBFSampler by glevv · Pull Request #24755 · scikit-learn/scikit-learn · GitHub
[go: up one dir, main page]

Skip to content

ENH Add gamma='scale' option to RBFSampler #24755

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.

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

Merged
merged 12 commits into from
Oct 28, 2022
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
4 changes: 4 additions & 0 deletions doc/whats_new/v1.2.rst
Original file line number Diff line number Diff line change
Expand Up @@ -350,6 +350,10 @@ Changelog
- |Enhancement| :class:`kernel_approximation.SkewedChi2Sampler` now preserves
dtype for `numpy.float32` inputs. :pr:`24350` by :user:`Rahil Parikh <rprkh>`.

- |Enhancement| :class:`kernel_approximation.RBFSampler` now accepts
`'scale'` option for parameter `gamma`.
:pr:`24755` by :user:`Gleb Levitski <GLevV>`

:mod:`sklearn.linear_model`
...........................

Expand Down
24 changes: 19 additions & 5 deletions sklearn/kernel_approximation.py
Original file line number Diff line number Diff line change
Expand Up @@ -249,8 +249,13 @@ class RBFSampler(ClassNamePrefixFeaturesOutMixin, TransformerMixin, BaseEstimato

Parameters
----------
gamma : float, default=1.0
gamma : 'scale' or float, default=1.0
Parameter of RBF kernel: exp(-gamma * x^2).
If ``gamma='scale'`` is passed then it uses
1 / (n_features * X.var()) as value of gamma.

.. versionadded:: 1.2
The option `"scale"` was added in 1.2.

n_components : int, default=100
Number of Monte Carlo samples per original feature.
Expand Down Expand Up @@ -319,7 +324,10 @@ class RBFSampler(ClassNamePrefixFeaturesOutMixin, TransformerMixin, BaseEstimato
"""

_parameter_constraints: dict = {
"gamma": [Interval(Real, 0, None, closed="left")],
"gamma": [
StrOptions({"scale"}),
Interval(Real, 0.0, None, closed="left"),
],
"n_components": [Interval(Integral, 1, None, closed="left")],
"random_state": ["random_state"],
}
Expand Down Expand Up @@ -354,8 +362,14 @@ def fit(self, X, y=None):
X = self._validate_data(X, accept_sparse="csr")
random_state = check_random_state(self.random_state)
n_features = X.shape[1]

self.random_weights_ = np.sqrt(2 * self.gamma) * random_state.normal(
sparse = sp.isspmatrix(X)
if self.gamma == "scale":
# var = E[X^2] - E[X]^2 if sparse
X_var = (X.multiply(X)).mean() - (X.mean()) ** 2 if sparse else X.var()
self._gamma = 1.0 / (n_features * X_var) if X_var != 0 else 1.0
else:
self._gamma = self.gamma
self.random_weights_ = (2.0 * self._gamma) ** 0.5 * random_state.normal(
size=(n_features, self.n_components)
)

Expand Down Expand Up @@ -390,7 +404,7 @@ def transform(self, X):
projection = safe_sparse_dot(X, self.random_weights_)
projection += self.random_offset_
np.cos(projection, projection)
projection *= np.sqrt(2.0) / np.sqrt(self.n_components)
projection *= (2.0 / self.n_components) ** 0.5
return projection

def _more_tags(self):
Expand Down
8 changes: 8 additions & 0 deletions sklearn/tests/test_kernel_approximation.py
Original file line number Diff line number Diff line change
Expand Up @@ -242,6 +242,14 @@ def test_rbf_sampler_dtype_equivalence():
assert_allclose(rbf32.random_weights_, rbf64.random_weights_)


def test_rbf_sampler_gamma_scale():
"""Check the inner value computed when `gamma='scale'`."""
X, y = [[0.0], [1.0]], [0, 1]
rbf = RBFSampler(gamma="scale")
rbf.fit(X, y)
assert rbf._gamma == pytest.approx(4)


def test_skewed_chi2_sampler_fitted_attributes_dtype(global_dtype):
"""Check that the fitted attributes are stored accordingly to the
data type of X."""
Expand Down
0