Paper available at: https://arxiv.org/abs/1811.05067

Peter Hase, John Benhart, Tianlin Duan, Liuyi Zhu, Cynthia Rudin

Duke Data Science Team

Instructions for Producing Sonnets

First install the required dependencies (given python 3.6x):

1. Tensorflow
2. gensim
3. numpy
4. argparse
5. nltk

Then download the 6 billion tokenn GloVe dictionary from [2] and unzip contents into storyline_for_reference

To train a model from the works of Walt Whitman, execute:

python train.py --data_dir data/whitman --save_dir whitman_model

Finally, execute the following to generate a poem using the Whitman model:

python genPoems.py topic seed

where topic is the user-supplied topic of the poem and seed (an optional argument) is an integer for the seed. We require the words in the topic to exist in the 6 billion token GloVe dictionary [2]. Generated poems can be found in the output_poems folder.