ml-playground

This repository contains a collection of small python projects implementing various ML models and applications. Each directory corresponds to an independent project with its own README with further information.

These include the following topics:

• Large Language Models (LLMs): Variety of examples and mini-projects, primarily using HuggingFace
• Diffusion Models: Coming soon...
• Graph Neural Networks (GNNs): Examples using PyG (PyTorch) and jraph (JAX)
• Reinforcement Learning (RL): Coming soon...
• Multi-layer perceptron (MLP) and logistic regression: Implementation of backpropagation from scratch

A variety of ML libraries are used across the different projects:

• PyTorch
• TensorFlow
• JAX

You can also find my code from my published research on ML applications to high-energy physics at the following:

• arXiv 2305.08979 : https://github.com/jdmulligan/JFN
• arXiv 2210.06450 : https://github.com/jdmulligan/ml-eic-flavor
• arXiv 2111.14589 : https://github.com/matplo/pyjetty/tree/master/pyjetty/alice_analysis/process/user/ml
• arXiv 2102.11337 : https://github.com/jdmulligan/bayesian-inference

And ongoing projects at:

• Conditional diffusion models: https://github.com/jdmulligan/QCD-Diffusion
• Graph neural networks: https://github.com/jdmulligan/Subjet-GNN

To set up a new project

The environment installation instructions for each project can be found in its corresponding README. (for organizational simplicity I elected to use separate environment/package management for each project within this git repo, since the projects will not be tagged/released).

In general, I use one of two different python package management tools, depending on the project: venv/pip or poetry.

venv / pip

Click for details

To initialize a new project:

mkdir MyProject
cd MyProject
mkdir myproject
python -m venv venv
source venv/bin/activate
pip install torch==2.0.0
...

And you can export the requirements with (or can write a setup.py):

pip freeze > requirements.txt

To install an existing project:

cd MyProject
python -m venv venv
source venv/bin/activate

To re-enter an existing environment, you can simply do:

cd MyProject
source venv/bin/activate

poetry

Click for details

To initialize a new project:

poetry new myproject
cd myproject
poetry shell
poetry add numpy
...
poetry install 

And you can export the requirements with:

poetry export -f requirements.txt --output requirements.txt --without-hashes

To install an existing project:

cd myproject
poetry install
poetry shell

To re-enter an existing environment, you can simply do:

cd myproject
poetry shell