philgookang / simple_bilstm_model

A simple many-to-one string generating Bidirectional LSTM using pytorch

Simple Bidirectional LSTM model

Genereate/predict a sentence using a Bidirectionl LSTM model.

Dependencies

For this simple code, we only need pytorch, numpy, and visdom. No other libraries required!
We use visdom to graph our loss value during training as seen in the picture below!

How To Use

1. Clone git

$ git clone https://github.com/philgookang/simple_bilstm_model.git
$ cd simple_bilstm_model

2. Train the model
To train the model yourself, run the following code. If you want to use a pretrained model, go direction to step 3.

$ python train.py

3. Test the model
To test the model, change the value for line 13 and run the following code.

$ python test.py

Loss graph

This is a picture of the loss value during training. While it may look very unclean, the dataset is only 20 sentences.
But it is converging if you look carefully! :)