deel.lip.constraints module¶
This module contains extra constraint objects. These object can be added as params to regular layers.
-
class
deel.lip.constraints.
AutoWeightClip
(scale=1)¶ Bases:
tensorflow.python.keras.constraints.Constraint
Clips the weights incident to each hidden unit to be inside the range [-c,+c]. With c = 1/sqrt(size(kernel)).
- Parameters
scale – scaling factor to increase/decrease clipping value.
-
get_config
()¶
-
class
deel.lip.constraints.
BjorckNormalizer
(niter_spectral=3, niter_bjorck=15, u=None)¶ Bases:
deel.lip.constraints.SpectralNormalizer
Ensure that all singular values of the weight matrix equals to 1. Computation based on BjorckNormalizer algorithm. The computation is done in two steps:
reduce the larget singular value to 1, using iterated power method.
increase other singular values to 1, using BjorckNormalizer algorithm.
- Parameters
niter_spectral – number of iteration to find the maximum singular value.
niter_bjorck – number of iteration with BjorckNormalizer algorithm..
u – vector used for iterated power method, can be set to None (used for serialization/deserialization purposes).
-
get_config
()¶
-
class
deel.lip.constraints.
FrobeniusNormalizer
(**kwargs)¶ Bases:
tensorflow.python.keras.constraints.Constraint
Clips the weights incident to each hidden unit to be inside the range [-c,+c]. With c = 1/norm(kernel).
-
class
deel.lip.constraints.
SpectralNormalizer
(niter_spectral=3, u=None)¶ Bases:
tensorflow.python.keras.constraints.Constraint
Ensure that the weights matrix have sigma_max == 1 (maximum singular value of the weights matrix).
- Parameters
niter_spectral – number of iteration to find the maximum singular value.
u – vector used for iterated power method, can be set to None (used for serialization/deserialization purposes).
-
get_config
()¶