trump_char_rnn

Character-Level RNN to Generate Speech in the Style of Donald Trump

Authors

Zayd Hammoudeh (zayd@ucsc.edu)
Ben Sherman (bcsherma@ucsc.edu)

Course

CMPS242 - Machine Learning
University of California - Santa Cruz
Prof. Manfred Warmuth
Fall 2017

Summary

Character-level recurrent neural network that generates text in the style of Donald Trump.

Running the Program

1. Python Version: 3.5.* or 3.6.*

• The code was specifically tested on 3.5.3 and 3.6.3 (Anaconda build).

2. Required Libraries

• TensorFlow
• Numpy
• pickle
• enum
• logging
• argparse

3. Training the Learner

• Run the file train.py. Default settings should be sufficient.
• It will automatically checkpoint the model every two epochs.
• We advise running at least 20 epochs (about 20 hours on our machine) to get an output of sufficient quality.

4. Generating the Text

• Run the file trump.py`. You must specify an input text seed and the character decision engine algorithm.
• An example entry is:
  python3 trump.py --seed "We will make America " --decision 11

Note: For more details on the parameters for both train.py and trump.py, you can run the program with the flag --help.