Torch Deterministic

Last release Python version Documentation Test status Test coverage Last commit

Torch Deterministic is a library for making PyTorch datasets with deterministic and easily reproducible augmentations. More specifically, this library is meant to facilitate the following paradigm:

  • None of the global PyTorch pseudorandom number generators (PRNGs) are used.

  • Every time the dataset is indexed, a new PRNG is seeded with that index number. This PRNG is used for all randomness needed to make the data point in question.