Skip to content

Latest commit

 

History

History

Generate TV Scripts

readme.md

Project: Generate TV Scripts

Introduction

In this project, you'll generate your own Seinfeld TV scripts using RNNs. You'll be using a Seinfeld dataset of scripts from 9 seasons. The Neural Network you'll build will generate a new, "fake" TV script.

Getting the project files

The project files can be found in our public GitHub repo, in the project-tv-script-generation folder. You can download the files from there, but it's better to clone the repository to your computer using:

git clone https://github.com/udacity/deep-learning-v2-pytorch.git

This way you can stay up to date with any changes we make by pulling the changes to your local repository with git pull.

You may also choose to complete your project using the provided, in-classroom project notebook; this will just require that you run the project notebook, and download the complete files, as specified below.

Review:

  • Consider using Python assertions for sanity testing - assertions are great for catching bugs. This is especially true of a dynamically type-checked language like Python where a wrong variable type or shape can cause errors at runtime
  • Logging is important for long-running applications. Logging done right produces a report that can be analyzed to debug errors and find crucial information. There could be different levels of logging or logging tags that can be used to filter messages most relevant to someone. Messages can be written to the terminal using print() or saved to file, for example using the Logger module. Sometimes it's worthwhile to catch and log exceptions during a long-running operation so that the operation itself is not aborted.
  • Check out this guide on debugging in python
  • Monitoring progress and debugging with Tensorboard: This tool can log detailed information about the model, data, hyperparameters, and more. Tensorboard can be used with Pytorch as well.
  • Profiling with Pytorch: Pytorch's profiler can be used to break down profiling information by operations (convolution, pooling, batch norm) and identify performance bottlenecks. The performance traces can be viewed in the browser itself. The profiler is a great tool for quickly comparing GPU vs CPU speedups for example.
  • You can check out more features of Dataloader here
  • Suggested Reading: Visual guide on LSTMs by Chris Olah: https://colah.github.io/posts/2015-08-Understanding-LSTMs/
  • Check out this wonderful guide on HyperParameter Optimization for Deep Neural Networks
  • These two guides explain the fundamentals of cross-entropy-loss beautifully: Visual Explanation of Binary Cross-Entropy Loss, Introduction to Cross-Entropy Loss