/LSTM

Character prediction using LSTM

Primary LanguagePythonMIT LicenseMIT

LSTM for character prediction

The task is to predict next character in the sequence using sequential model like LSTM . Here i have used shakespeare dataset to see how well does LSTM composes its own writing using some context learn over time on shakespeare writing.

How to run the code:

#To test without peephole connections.
python3 code/lstm_without_peephole.py
#To test it with peephole connections.	
python3 code/lstm_with_peephole_connections.py
#To test Bi-directional LSTM.	
python3 code/LSTM/blstm_train.py

Loss graph of LSTM without Peephole connection:

 LSTM without Peephole connection

Loss graph of LSTM with Peephole connection:

 LSTM with Peephole connection

Loss graph of Bi-directional LSTM with Peephole connection:

 BLSTM with Peephole connection

Things Completed:

  • LSTM without peephole connections.
  • LSTM with peephole connections.
  • Bidiectional LSTM.
  • Nested LSTM.
  • GRU Unit.
  • Support for custom loss Function.