What It Does. Like pytest supports you in writing better Python code through automated testing, pytrain helps you build differentiable programs by making it easy to create tasks and use them to optimize your programs.

How It Works. You write differentiable components (e.g. using PyTorch) along with tasks they are expected to handle, and pytrain will optimize all the parameters involved and save them to disk as an automated process.

# Create a base environment
conda create -n myenv python=3.6
conda install pytorch -c pytorch

# Either install latest release 0.0.x, see GitHub for latest version number:
pip install https://github.com/alexjc/pytrain/releases/download/v0.0.x/pytrain-0.0.x.tar.gz

# Or clone the repository online:
git clone https://github.com/alexjc/pytrain.git

# Either launch from installed script:
pytrain -h
pytrain --path examples/

# Or run from current directory:
python -m pytrain -h
python -m pytrain --path examples/

See the #docs/ folder or scripts in #examples/ to get up and running.

NOTE: This version 0.0.x is an early prototype for the PyTorch Hackathon 2019. Feedback and suggestions are the most welcome at this stage!