char-rnn

A Keras implementation of Andrej Karpathy's famous RNN blog post. These python scripts can be used to

• Train a deep LSTM-RNN on textual data
• Generate new data based on a seed
• Visualize training growth

Prerequisites

This repo is built in python3. The following frameworks are used:

• NumPy
• TensorFlow
• Theano
• Keras

With the following python modules:

• sklearn
• matplotlib

Usage

To train a new RNN model on a text file TRAIN_FILE:

python3 rnn.py -i TRAIN_FILE -e NUM_EPOCHS

In order to resume training from previous session, include the -c option and add -w WEIGHTS_FILE containing your previous weights file.

Files

rnn.py # Module for training library
generate.py # Module for generating arbitrary text
graph_training.py # Module for graphing the training history generated by rnn.py
callback.py # Helper function to generate text as training progresses.

Authors

• Michael Seaman - Initial work - MichaelSeaman

License

This project is licensed under the MIT License.

Acknowledgments

• Andrej Karpathy
• Jason Brownlee
• Keras Tutorials