troiwill / torch-mdn

A set of functions for building and evaluating a Mixture Density Network in PyTorch.

PyTorch MDN

PyTorch MDN (torch-mdn) is a set of classes and functions for building and evaluating a Mixture Density Network in PyTorch.

Setting up the environment

Run the following commands to set up the environment.

cd ${HOME}
mkdir -p repos && cd repos
git clone https://github.com/troiwill/torch-mdn.git
conda create -n torch-mdn python=3.8 pytorch==1.10.2 matplotlib

# Activate the conda environment and install Python dependencies.
conda activate torch-mdn
pip install pydantic==1.10.10

Installing the torch_mdn Python package

Installing this Python package requires a two-step process. First, you must build the package. Assuming you set up the environment as mentioned above, run the following commands:

conda activate torch_mdn
cd ${HOME}/repos/torch-mdn
python -m build

Once you built the package, use pip to install the wheel file (*.whl). For example, pip install <TORCH_MDN_WHEEL>.whl. If you do not want to install this package in the current conda environment, deactivate the environment first (using conda deactivate) and then install the package in the appropriate environment.

Testing the Library

We use the pytest package to sanity check the components in the library. To install pytest, run:

pip install pytest

To run all the tests, run:

cd tests
pytest

To run individual test files, run:

cd tests
pytest test_<file name>.py # for example, test_utils.py