patrickvossler18 / csl

csl: PyTorch-based Constrained Learning

csl

Learning under requirements with pytorch

• Documentation: https://csl.readthedocs.io/en/latest/

What is it?

csl (standing for Constrained Statistical Learning) is a Python package based around pytorch to simplify the definition of constrained learning problems and then solving them.

It was developed to run experiments for my research on learning under requirements.

Requirements

• numpy
• pytorch
• matplotlib (for plotting)
• pandas (only for csl.datasets)
• PIL (only for csl.datasets)

Installation

In your working folder simply do

   $ git clone https://github.com/lchamon/csl.git

or download and extract.

If you use conda, you can set up a ready-to-go requirements by running

   $ conda env create -f environment.yml
   $ conda activate csl

Note: This environment uses pytorch without GPU support. If you need GPU support, you should replace the package cpuonly in environment.yml with cudatoolkit=XX.X where XX.X denotes your CUDA version.

License

csl is distributed under the MIT license, see LICENSE.