callback module

DIRESA callback classes/functions

Author:

Geert De Paepe

Email:

geert.de.paepe@vub.be

License:

MIT License

class callback.LossWeightAnnealing(*args: Any, **kwargs: Any)

https://keras.io/guides/writing_your_own_callbacks/ https://medium.com/dive-into-ml-ai/adaptive-weighing-of-loss-functions-for-multiple-output-keras-models-71a1b0aca66e

__init__(weight, loss_name, target_loss=3e-06, anneal_step=0.1, start_epoch=5)

Parameters:

• weight – keras.backend.variable with initial loss weight

• loss_name – name of the loss function to apply the annealing

• target_loss – target loss, weight is increased until loss < target

• anneal_step – annealing step size for increasing weight

• start_epoch – epoch from which annealing starts

on_epoch_end(epoch, logs=None)

Executed during training on each epoch end Increases weight with anneal_step as from start_epoch until unweighted loss is smaller than target Prints loss weight, reconstruction loss and unweighted loss :param epoch: number of epoch :param logs: dict containing the loss values

on_epoch_begin(epoch, logs=None)