davissorenson / vscode-pytorch

Visual Studio Code (VS Code) extension offering code snippets when working with PyTorch or Fastai.

PyTorch Code Snippets for VSCode

This project aims to provide a faster workflow when using the PyTorch or fastai library in Visual Studio Code. This extension provides code snippets for often used coding blocks as well as code example provided by the libraries for common deep learning tasks.

Table of Contents

1. Getting Started
   1. PyTorch Preview
   2. Fastai Preview
2. Dependencies
3. Installation
4. Changelog
5. Get Involved
   1. Contributing
   2. Future Plans
   3. Contact
6. Credits

Getting Started

If this extension is installed and activated you might notice by start typing pytorch or fastai in your python file a dropdown list appears. Entries starting with pytorch: or fastai: represent code snippets refering to those libraries offered by this extension. Snippets appear in the same way as code completion by using Ctrl+Space. There are placeholders, variables and choices defined inside the code snippets so you can jump from one to another just by pressing Tab. Examples on how to use this extension are shown below. For a full overview please take a look at the Snippet Collection.

PyTorch Preview

Fastai Preview

Dependencies

Currently the snippets are working for the latest releases of the supported libraries:

• fastai >= 1.0.21
• pytorch >= 0.4.1

Info: Be aware that at the time the fastai library is changing rapidly, so I can not guarantee that every snippet is actually working. I will try to keep the snippets in line with the latest fastai library developments (as well as the latest PyTorch releases), so it might not work for you in case you are using an outdated version.

Installation

The installation process is very straightforward. The recommend, stable way is as follows:

1. Launch Visual Studio Code
2. From the command palette Ctrl-Shift-P (Windows, Linux) or Cmd-Shift-P (MacOS) select Install Extension or just simply go to the Extensions tab (most likely the fifth entry on the left sidebar). You can also use the hotkey Ctrl-Shift-X respectifely Cmd-Shift-X
3. Type PyTorch Snippets
4. Choose and install extension
5. Restart/reload Visual Studio Code

This will give you the most recent version you can find on the VS Marketplace. Alternatively you can also clone this repository and move it manually into your VS Code extension folder which will give you the most recent version on GitHub.

git clone https://github.com/SvenBecker/vscode-pytorch.git
mv vscode-pytorch /path/to/your/VSCodeExtensionFolder/

On Windows for example you can normally find the extension folder at C:\Users\YourName\.vscode\extensions. On MacOS and Linux it should be located at ~/.vscode/extensions.

Info: The second method will give you the overall most recent version because I won't update the VS Marketplace version as often as the GitHub version but the VS Marketplace version will probably be the more stable one.

Changelog

Version 0.2.1:

Added

• Added PyTorch Functional Snippets pytorch:F:

Version 0.2.0:

Added

• Added code snippets for fast metrics or loss selection (PyTorch and fastai)
• Added easy selection of neural network layers in PyTorch based on their type pytorch:layer: (conv, recurrent, etc.)
• Added some more PyTorch snippets like for example optimizer selection, weight initialization etc.
• Added DataBlock API snippets for fastai

Changes

• Changed some code examples for common problems to be inline with the official examples provided by PyTorch

All of the past as well as the upcoming changes can further be viewed at Changelog.

Get Involved

Contributing

If you want to contribute, what I would highly appreciate since this project is currently in a very early stage and there is still so much to do, please take a look at Contributing.

Future Plans

I would like to extend snippet options to additional packages which are related to PyTorch. This might include Ignite or Pyro.

Furthermore, but not in this project, I would like to offer VS Code snippets for visualization in Python, like for example Matplotlib or Seaborn. Eventually even considering the inclusion of interactive visualization libraries like Bokeh or Plotly. If anyone is interested in this kind of project and/or want to offer some help please contact me.

Contact

Suggestions for improvements will be highly appreciated. You can write me an email (address is provided on my profile) or you can contact me via Twitter @SBX9209.

Credits

• PyTorch: A huge thanks to the PyTorch team for building such an awesome, easy to use deep learning framework and for offering a numerous amount of code examples.
• Fastai: Also a huge thanks to Jeremy Howard and the whole fast.ai team for creating this great deep learning library and also for offering wonderful, very practical orientated free lectures giving insight to deep learning to a wide range of people including ml/dl beginners as well as advanced practitioners.
• VSCode: My personal favorite text editor and I'm grateful for all the options they are offering, like a huge amount of language support, extensions, integrated terminal, debugger etc..