Deep Learning with PyTorch and a Vim Based IDE

Installation Instructions

0. Build and tag the deep learning image.

git clone https://github.com/vim-ide-deep-learning-docker-image
cd vim-ide-deep-learning-docker-image/deep-learning
docker build -t deep-learning .
cd ..

1. (Optional, allows TabNine to use your settings and deep learning model) Clone this repository and add a copy of your TabNine configuration folder into this directory (on Linux this is ~/.config/TabNine). If you do not have TabNine (consider rectifying this unfortunate reality) comment out the TabNine configuration in the vim-ide Docker file.

cp -R ~/.config/TabNine deep-learning-docker/TabNine vim-ide
cp ~/.local/share/TabNine/models/model-name vim-ide

2. Build the vim-ide image using the deep-learning image as the base:

cd vim-ide
docker build -t vim-ide .

3. Run the image in interactive mode:

docker run -it --gpus all vim-ide

Details

The deep-learning image is based on NVIDIA's PyTorch image and includes CUDA 10 and associated optimization libraries, Anaconda with the SciPy stack, and of course PyTorch. Also included are some useful Python libraries:

• ax-platform for Bayesian optimization.
• ray for distributed hyperparameter tuning and logging.
• mlflow for experiment tracking (integrates easily with ray).
• dask for distributed data wrangling.
• awscli and boto3 for AWS.
• tqdm for progress bars.

The Vim image includes a full featureed IDE with nerdtree for file browsing, PEP 8 formatting, code folding, ctags for searching, TabNine for AI powered code completion, solarized as the theme, and syntastic for catching silly bugs (I disable warnings). Also:

• black for formatting.
• flake8 with black compatability.

I built this image to provide a full featured IDE capable of running via ssh. The original intention was to replace PyCharm and the remote deploy feature, but now I like it better thanks to the total removal of the mouse. Check out vast.ai for a cheap peer-to-peer cloud option that runs the Docker image you specify.

Useful Commands in Vim

• C(control)-n opens the nerdtree file browser.
• When the cursor is on a Python function, C-W ] splits the window and opens the source file.
• F5 toggles the dark and light solarized themes.
• Tab completes suggestions.