nntoolbox.callbacks.bptt module

class nntoolbox.callbacks.bptt.VariableLengthBPTT(default_len: int, p: float, std: float)

Bases: nntoolbox.callbacks.callbacks.Callback

Change the truncated backprop through time length and linearly scale the learning rate. (UNTESTED)

References:

Stephen Merity, Nitish Shirish Keskar, Richard Socher. “Regularizing and Optimizing LSTM Language Models.” https://arxiv.org/abs/1708.02182

on_epoch_begin()

on_epoch_end(logs: Dict[str, Any]) → bool