Skip to content

Allow disable covariance update in GP layer. #456

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Allow disable covariance update in GP layer. #456

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
34 changes: 23 additions & 11 deletions edward2/tensorflow/layers/random_feature.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,12 +70,13 @@ def __init__(self,
gp_cov_momentum=0.999,
gp_cov_ridge_penalty=1e-6,
scale_random_features=True,
return_random_features=False,
use_custom_random_features=True,
custom_random_features_initializer=None,
custom_random_features_activation=None,
l2_regularization=0.,
gp_cov_likelihood='gaussian',
return_gp_cov=True,
return_random_features=False,
dtype=None,
name='random_feature_gaussian_process',
**gp_output_kwargs):
Expand Down Expand Up @@ -103,7 +104,6 @@ def __init__(self,
covariance matrix.
scale_random_features: (bool) Whether to scale the random feature
by sqrt(2. / num_inducing).
return_random_features: (bool) Whether to also return random features.
use_custom_random_features: (bool) Whether to use custom random
features implemented using tf.keras.layers.Dense.
custom_random_features_initializer: (str or callable) Initializer for
Expand All @@ -116,6 +116,9 @@ def __init__(self,
weights.
gp_cov_likelihood: (string) Likelihood to use for computing Laplace
approximation for covariance matrix. Default to `gaussian`.
return_gp_cov: (bool) Whether to also return GP covariance matrix.
If False then no covariance learning is performed.
return_random_features: (bool) Whether to also return random features.
dtype: (tf.DType) Input data type.
name: (string) Layer name.
**gp_output_kwargs: Additional keyword arguments to dense output layer.
Expand All @@ -130,6 +133,7 @@ def __init__(self,

self.scale_random_features = scale_random_features
self.return_random_features = return_random_features
self.return_gp_cov = return_gp_cov

self.gp_kernel_type = gp_kernel_type
self.gp_kernel_scale = gp_kernel_scale
Expand Down Expand Up @@ -168,12 +172,14 @@ def build(self, input_shape):
self._random_feature = self._make_random_feature_layer(
name='gp_random_feature')

self._gp_cov_layer = self.covariance_layer(
momentum=self.gp_cov_momentum,
ridge_penalty=self.gp_cov_ridge_penalty,
likelihood=self.gp_cov_likelihood,
dtype=self.dtype,
name='gp_covariance')
if self.return_gp_cov:
self._gp_cov_layer = self.covariance_layer(
momentum=self.gp_cov_momentum,
ridge_penalty=self.gp_cov_ridge_penalty,
likelihood=self.gp_cov_likelihood,
dtype=self.dtype,
name='gp_covariance')

self._gp_output_layer = self.dense_layer(
units=self.units,
use_bias=False,
Expand Down Expand Up @@ -253,11 +259,17 @@ def call(self, inputs, global_step=None, training=None):

# Computes posterior center (i.e., MAP estimate) and variance.
gp_output = self._gp_output_layer(gp_feature) + self._gp_output_bias
gp_covmat = self._gp_cov_layer(gp_feature, gp_output, training)

if self.return_gp_cov:
gp_covmat = self._gp_cov_layer(gp_feature, gp_output, training)

model_output = [gp_output,]
if self.return_gp_cov:
model_output.append(gp_covmat)
if self.return_random_features:
return gp_output, gp_covmat, gp_feature
return gp_output, gp_covmat
model_output.append(gp_feature)

return model_output


class LaplaceRandomFeatureCovariance(tf.keras.layers.Layer):
Expand Down