I am consolidating my PyTorch work for PhD into one single repository of installable packages. I want my work to be easily reproducible and transparent (plus easy) to review. You may find that some of these are rehash of tried and tested techniques, but I subscribe to the idea of "what I cannot create, I don't understand". Some of the (upcoming) work packages here include:
- Continual Machine Learning
- Spiking Neural Network
Over the years, I have also found several utilities and design patterns to be ubiquitous across different PyTorch projects. I want to put them into a utilities library for easy use in future, instead of always looking up such patterns in documentation, tutorials, or stackoverflow. The patterns package is my attempt to do so.
Tests may be a little overkill for non-production code, but it may be a good idea to add them in future.
Installation:
git clone https://github.com/danqiye1/phd-lib
cd phd-lib
pip install -e .There are currently 2 packages:
continual: Tools and experiments on continual learning
patterns: Tools for normal experiments
Experiments can be rerun as package modules:
# Example for running rehearsal
python -m continual.rehearsal