Skip to content

Latest commit

 

History

History
51 lines (42 loc) · 3.58 KB

CONTRIBUTING.md

File metadata and controls

51 lines (42 loc) · 3.58 KB
Raw

Welcome to the contributing guidelines !

Want to contribute to pythae library ? That is cool! Thank you! 😄

Contributing guidelines

If you want to contribute to this repo, please consider following this checklist

  1. Fork this repo
  2. Clone it on your local drive and install the library in editable mode
$ git clone git@github.com:your_github_name/benchmark_VAE.git
$ cd benchmark_VAE
$ pip install -e .
  1. Create a branch with an explicit name of your contribution
$ git checkout -b my_branch_with_contribution

  1. Make sure you add the appropriate tests to test your feature. If the library test coverage reduces significantly, the contribution will raise some red flags.

  2. Ensure that your contribution passes the existing test suite by running

pytest tests/

  • Polish your contribution using black and isort

  • Finally, open a pull request directly on your Github ! 🚀

Implementing a new model

If you want to add a new model please make sure that you followed the following checklist:

  • Create a folder named your_model_name in pythae/models and containg a file with the model implementation entitled your_model_name_model.py and a file with the model configuration named your_model_name_config.py.
  • The your_model_name_model.py file contains a class with the name of your model inheriting from either AE or VAE classes depending on the model architecture.
  • The your_model_name_config.py files contains a dataclass inheriting from either AEConfig or VAEConfig. See for instance pythae/models/rae_l2 folder for a AE-based models and pythae/models/rhvae folder for a VAE-based models
  • The model must have a forward method in which the loss is computed and returning a ModelOutput instance with the loss being stored under theloss key.
  • You also implemented the classmethods load_from_folder and _load_model_config_from_folder allowing to reload the model from a folder. See pythae/models/rae_l2 for instance.
  • Make your tests in the tests folder. See for instance pythae/tests/test_rae_l2_tests.py. You will see that the tests for the models look the same and cover them quite well. Hence, you can reuse this test suite as an inspiration to test your model.

Implementing a new sampler

If you want to add a new sampler please make sure that you followed the following checklist:

  • Create a folder named your_sampler_name in pythae/samplers and containg a file with the sampler implementation entitled your_sampler_name_sampler.py and a file with the sampler configuration (if needed) named your_sampler_name_config.py. See pythae/samplers/gaussian_mixture for instance.
  • The your_sampler_name_sampler.py files contains a class with the name of yoyr sampler inheriting from BaseSampler.
  • The your_sampler_name_config.py files contains a dataclass inheriting from BaseSamplerConfig. See pythae/samplers/gaussian_mixture/gaussian_mixture_config.py.
  • The sampler must have a sample method able to save the generated images in a folder and return them if desired.
  • If the sampler needs to be fitted, a fit merhod can be implemented. See pythae/samplers/gaussian_mixture/gaussian_mixture_samplers.py for example.
  • Make your tests in the tests folder. See for instance pythae/tests/test_gaussian_mixture_sampler.py. You will see that the tests for the samplers look the same, you can reuse this test suite as an inspiration to test your sampler.

Any doubts ?

In any case if you have any question, do not hesitate to reach out to me directly, I will be happy to help! 😄