Word level and character level text generation using RNN

Overview

An LSTM model was trained on five parts of Harry Potter series to produce new text whose distribution resembles the original text.

Dependencies

• Python 3.6
• Pytorch
• Bcolz
• PyPDF2

Installation

All the dependency packages can be installed with pip install command. If anaconda distribution is installed then conda install can be used.

Data and pre processing

Five books from Harry Potter series were pre processed and made into a text file such that each line contains a sentence that can be given as an input to the LSTM network.

Training

Both the word level and character level model can be trained by running python train.py from the terminal. The training process also outputs the sampled text after the mentioned iteration interval.