README.md

Training an RNN model

Session A: Data Collection and Python Environment

Objectives

• Learn about preparing a text dataset
• Learn how to setup a native TensorFlow Python local environment

Local Python Environment

• Python and Virtualenv Tutorial
• Creation of Virtual Environments Documentation
• ml5 charRNN training repo
• Training Steps

Running p5.js sketches locally

• node.js
• git
• Visual Studio Code
• All Coding Train Workflow Videos

ml5.js examples

• charRNN Generate Text
• charRNN Interactive Text
• charRNN "stateful" Text

Assignment:

1. Collect a text dataset minimum 5 MB of text. We'll use this dataset in class on Wednesday.

Session B: Deploying the Model

Objectives

• Review and continue to learn about training a charRNN model.
• Understand how “temperature” affects the charRNN’s generated text.
• Understand the distinction between “stateful” and “stateless” generation.

Cloud Computing

• Spell
  • Introduction to Spell
  • Text Generation with LSTM and Spell with Nabil Hassein
• NYU HPC
  • HPC Tutorial and Documentation by Dan Oved
• Google Colab
• AWS
• Paperspace

Text Data Sources

• Project Gutenberg

GitHub Pages

• GitHub Pages documentation
• GitHub Pages Video Tutorial -- this is from 2016, several things have changed, the most significant being that you no longer are required to use gh-pages as the branch name.

Assignment (Due Sunday Oct 27, 6pm):

Reading / Viewing

• Eyeo 2017 Talk about Writing with the machine by Robin Sloan (Text & Music)
• The Subtext of a Black Corpus In conversation with ITP SIRs Nikita Huggins & Ayodamola Okunseinde by Ashley Lewis
• Adventures in Narrated Reality Part 1 and Adventures in Narrated Reality Part 2 by Ross Goodwin

Instructions

We didn't get as far as I would have liked in class so some of the steps below may prove difficult for you. Please reach out if you are having trouble. At a minimum, complete steps 1 and 2 and make an attempt at Step 3. If you run into trouble, document the process in your post.

1. Collect a text dataset. Try to achieve a minimum of 5 MB of text.
2. Reflect on the readings above. How did they inform your choice of text data and collection process?
3. Set up a local or cloud environment and train a model with your own data.
4. Create an interactive sketch with your model and ml5.js. (Note that you will not be able to run your p5.js sketch in the editor since it does not support uploading of charRNN models. You can run your sketch locally and document the results on your blog. Include your code (an easy way is to link to a GitHub Gist. If you are feeling ambitious I would suggest hosting your sketch via GitHub Pages.)
5. Link your blog post and any related materials on the Assignment 8 Wiki.