Skip to content

Latest commit

 

History

History
9 lines (6 loc) · 629 Bytes

README.md

File metadata and controls

9 lines (6 loc) · 629 Bytes

LSTM-Numpy

Description

The implementation of the LSTM network from scratch using Numpy in Python, and the training of a simple word level language model.

This code is an implementation of the LSTM variant called CIFG (Coupled Input and Forget Gates) from the paper LSTM : A Search Space Odyssey link

The model is a two stacked layers of LSTMs followed by a linear classification layer for decoding the next character. It is trained on a shakespeare corpus.

The code also contains a simple implementation of the SGD with momentum algorithm for the network parameters optimization.