This package contains mainly loss functions, model definitions and metrics in both functional and modular and (whenever possible) pure PyTorch implementations.
pip install git+https://github.com/justusschock/dl-utils
pip install deep-learning-utils
Currently there are the following subpackages:
dlutils.data
: contains data utilities (so far just a dataset for random fake data)dlutils.losses
: extends the losses given in PyTorch itself by a few more loss functionsdlutils.metrics
: implements some common metricsdlutils.models
: contains Nd implementations of many popular modelsdlutils.models.gans
: contains many basic gan implementations, but so far not for arbitrary dimensions
dlutils.optims
: containis additional optimizersdlutils.utils
: contains additional utilities such as tensor operations and module loading
-
Most of this code was only tested sparely and not with a proper CI/CD and unittests. I'm currently working on that and any contributions are highly welcomed.
-
All implementations are done for pure PyTorch. You can employ them in whatever training framework you want (like [pytorch/ignite]{https://github.com/pytorch/ignite) or Pytorch-Lightning) or in your custom training loops