mdml-tools Python Library
This python library was created with the purpose of collecting all kind of boilerplate code that I (and some of my colleagues) frequently use in our projects. We realized that we often copy and paste code from one project to another, especially hydra dataclasses. This library does not only contain these dataclasses it also contains a script to automatically create hydra dataclasses from python modules, classes and functions. This makes it very easy to work with hydra and to create a clean and structured project.
Furthermore, all kind of utilites are collected in this library. For example, a profiler that can be used to profile
contexts or functions in torch.nn modules. Also very helpful are the standard-normalzation-scaler and the
min-max-scaler in pure pytorch. They provide the same functionality as their SciPy counterparts but are fully
differentiable. The last utility I want to higlight here is the possibility to read data from tensorboard logs with a
single function call. This makes it very easy to extract the data from tensorboard to e.g. plot it with matplotlib for
implementation in a publication.