Toolkit, containing modules and helper methods for machine learning projects. This package contains any modules and helper functions that I have used repeatedly in my projects. I hope it will prove useful to others as well. The main package contains very general functionality as well as sub-packages tailored for specific operations (e.g. neural net modules, filesystem operations, etc.). These subpackages are outlined in the following sections.
The filesystem sub-package has functionality related to file operations. For example if you need to list all files in a directory you can do this with:
import dtk.filesystem as dfs
files = dfs.list_files("path")
If you have multiple directories with matching files (i.e. for storing corresponding data from different domains). You can find all the matching files by doing:
import dtk.filesystem as dfs
directories=["dir1", "dir2", "dir3"]
files = dfs.list_matching_files(directories)
The media sub-package has functionality for handling media. With the methods in the media subpackage you can save media or convert to bytestream (for logging with Weights&Biases).
The nn sub-package has useful building blocks that can be used to construct deep neural nets as well as losses. It builds on top of Pytorch
The transforms sub-package contains useful transformations (mainly for sequential data)