Module: tfl.pwl_calibration_layer

View source on GitHub

Piecewise linear calibration layer.

Keras implementation of tensorflow lattice pwl calibration layer. Layer takes single or multi-dimensional input and transforms it using piecewise linear functions following monotonicity, convexity/concavity and bounds constraints if specified.

Classes

class HessianRegularizer: Hessian regularizer for PWL calibration layer.

class LaplacianRegularizer: Laplacian regularizer for PWL calibration layer.

class NaiveBoundsConstraints: Naively clips all elements of tensor to be within bounds.

class PWLCalibrationConstraints: Monotonicity and bounds constraints for PWL calibration layer.

class UniformOutputInitializer: Initializes PWL calibration layer to represent linear function.

class WrinkleRegularizer: Wrinkle regularizer for PWL calibration layer.

Other Members

  • INTERPOLATION_KEYPOINTS_NAME = 'interpolation_keypoints'
  • LENGTHS_NAME = 'lengths'
  • MISSING_INPUT_VALUE_NAME = 'missing_input_value'
  • PWL_CALIBRATION_KERNEL_NAME = 'pwl_calibration_kernel'
  • PWL_CALIBRATION_MISSING_OUTPUT_NAME = 'pwl_calibration_missing_output'