lbfgsTrainSteps

Training step and functions for training with the L-BFGS optimizer.

This is implementation was adapted from a method for using L-BFGS in learning to function match with a keras sequential network, developed by Pi-Yueh Chuang.

value_and_gradients_generator(model, lossfn, lossfn_data, eqns, extras)

Generator which returns a function to compute the value_and_graidents function in tfp.optimizer.lbfgs_minimize

Parameters:
  • model (model) –

    TensorFlow model to train.

  • lossfn (function) –

    Loss function taken from other train steps to minimize

  • lossfn_data (list) –

    List of data required for loss function.

  • eqns (list) –

    List of equations to learn.

  • extras (list) –

    List of extra data required for loss function depending on trainign routine.
Returns:
  • function –

    Value and gradients function.

lbfgsTrain(model, eqns, epochs, lossfn, lossfn_data, extras)

Initiaalizes training for tfp.optimizer.lbfgs_minimize training

Parameters:
  • model (model) –

    TensorFlow model to train.

  • eqns (list) –

    List of equations to learn.

  • epochs (int) –

    Outer iterations for L-BFGS to take.

  • lossfn (function) –

    Loss function taken from other train steps to minimize

  • lossfn_data (list) –

    List of data required for loss function.

  • extras (list) –

    List of extra data required for loss function depending on trainign routine.
Returns:
  • list –

    Trainign loss of inner iterations.

  • int –

    Number of inner training iterations.