Callback for RNN training

Callback that uses the outputs of language models to add AR and TAR regularization

ModelResetter


def ModelResetter(
    after_create:NoneType=None, before_fit:NoneType=None, before_epoch:NoneType=None, before_train:NoneType=None,
    before_batch:NoneType=None, after_pred:NoneType=None, after_loss:NoneType=None, before_backward:NoneType=None,
    after_cancel_backward:NoneType=None, after_backward:NoneType=None, before_step:NoneType=None,
    after_cancel_step:NoneType=None, after_step:NoneType=None, after_cancel_batch:NoneType=None,
    after_batch:NoneType=None, after_cancel_train:NoneType=None, after_train:NoneType=None,
    before_validate:NoneType=None, after_cancel_validate:NoneType=None, after_validate:NoneType=None,
    after_cancel_epoch:NoneType=None, after_epoch:NoneType=None, after_cancel_fit:NoneType=None,
    after_fit:NoneType=None
):

Callback that resets the model at each validation/training step

source

RNNCallback


def RNNCallback(
    after_create:NoneType=None, before_fit:NoneType=None, before_epoch:NoneType=None, before_train:NoneType=None,
    before_batch:NoneType=None, after_pred:NoneType=None, after_loss:NoneType=None, before_backward:NoneType=None,
    after_cancel_backward:NoneType=None, after_backward:NoneType=None, before_step:NoneType=None,
    after_cancel_step:NoneType=None, after_step:NoneType=None, after_cancel_batch:NoneType=None,
    after_batch:NoneType=None, after_cancel_train:NoneType=None, after_train:NoneType=None,
    before_validate:NoneType=None, after_cancel_validate:NoneType=None, after_validate:NoneType=None,
    after_cancel_epoch:NoneType=None, after_epoch:NoneType=None, after_cancel_fit:NoneType=None,
    after_fit:NoneType=None
):

Save the raw and dropped-out outputs and only keep the true output for loss computation

source

RNNRegularizer


def RNNRegularizer(
    alpha:float=0.0, beta:float=0.0
):

Add AR and TAR regularization

source

rnn_cbs


def rnn_cbs(
    alpha:float=0.0, beta:float=0.0
):

All callbacks needed for (optionally regularized) RNN training