Thanks for tuning in to Google I/O. View all sessions on demandWatch on demand

Module: tfl.kronecker_factored_lattice_layer

Kronecker-Factored Lattice layer with monotonicity constraints.

Keras implementation of tensorflow Kronecker-Factored Lattice layer. This layer takes one or more d-dimensional input(s) and combines them using a Kronecker-Factored Lattice function, satisfying monotonicity constraints if specified.


class BiasInitializer: Initializes bias depending on output_min and output_max.

class KFLRandomMonotonicInitializer: Initializes a tfl.layers.KroneckerFactoredLattice as random monotonic.

class KroneckerFactoredLatticeConstraints: Constraints for tfl.layers.KroneckerFactoredLattice layer.

class ScaleConstraints: Constraints for tfl.layers.KroneckerFactoredLattice scale.

class ScaleInitializer: Initializes scale depending on output_min and output_max.


create_kernel_initializer(...): Returns a kernel Keras initializer object from its id.

create_scale_initializer(...): Returns a scale Keras initializer object from its id.

DIMS_NAME 'dims'
KFL_BIAS_NAME 'kronecker_factored_lattice_bias'
KFL_KERNEL_NAME 'kronecker_factored_lattice_kernel'
KFL_SCALE_NAME 'kronecker_factored_lattice_scale'
LATTICE_SIZES_NAME 'lattice_sizes'
NUM_TERMS_NAME 'num_terms'
UNITS_NAME 'units'
absolute_import Instance of __future__._Feature
division Instance of __future__._Feature
print_function Instance of __future__._Feature