tfl.lattice_layer.LatticeConstraints

View source on GitHub

Class LatticeConstraints

Constraints for tfl.layers.Lattice layer.

Applies monotonicity, unimodality, trust and bound constraints to the lattice parameters. See tfl.layers.Lattice for details.

Attributes:

  • All __init__ arguments.

__init__

View source

__init__(
    lattice_sizes,
    monotonicities=None,
    unimodalities=None,
    edgeworth_trusts=None,
    trapezoid_trusts=None,
    monotonic_dominances=None,
    joint_monotonicities=None,
    output_min=None,
    output_max=None,
    num_projection_iterations=1,
    enforce_strict_monotonicity=True
)

Initializes an instance of LatticeConstraints.

Args:

  • lattice_sizes: Lattice sizes of Lattice layer to constraint.
  • monotonicities: Same meaning as corresponding parameter of Lattice.
  • unimodalities: Same meaning as corresponding parameter of Lattice.
  • edgeworth_trusts: Same meaning as corresponding parameter of Lattice.
  • trapezoid_trusts: Same meaning as corresponding parameter of Lattice.
  • monotonic_dominances: Same meaning as corresponding parameter of Lattice.
  • joint_monotonicities: Same meaning as corresponding parameter of Lattice.
  • output_min: Minimum possible output.
  • output_max: Maximum possible output.
  • num_projection_iterations: Same meaning as corresponding parameter of Lattice.
  • enforce_strict_monotonicity: Whether to use approximate projection to ensure that constratins are strictly satisfied.

Raises:

  • ValueError: If weights to project don't correspond to lattice_sizes.

Methods

__call__

View source

__call__(w)

Applies constraints to w.

get_config

View source

get_config()

Standard Keras config for serialization.