Name: Nathan
Type: User
Company: @dragonfruit-ai
Bio: ML engineer with interests in representation learning, computer vision, high performance computing, reinforcement learning, generative art, and clean code
Location: Johannesburg, South Africa
Blog: michlo.dev
Nathan's Projects
My research on improving Population Based Training.
[WIP] ๐จ Create and convert machine learning datasets
๐งถ Modular VAE disentanglement framework for python built with PyTorch Lightning โธ Including metrics and datasets โธ With strongly supervised, weakly supervised and unsupervised methods โธ Easily configured and run with Hydra config โธ Inspired by disentanglement_lib
๐ช Essential utilities for working with files. Including easy downloading, fast hashing, stale file detection, dataset sharding, proxy downloading, file renaming
๐ A sane but flexible configuration framework inspired by Hydra config, with yaml and pythonic backends.
Variance normalising pre-training of neural networks.
WIP ๐งโ๐พ Minimal Reinforcement Learning Framework
A Github Action to tag a repo on merge.
๐จโ๏ธ๐ Resources for "Overlooked Implications of the Reconstruction Loss for VAE Disentanglement" submitted to IJCAI 2023
Toolbox of models, callbacks, and datasets for AI/ML researchers.
๐ฆ โก๏ธ Neural Cellular Automata (NCA) implemented with PyTorch Lightning
๐ Repository containing all the research for my MSc. "Disentanglement Using VAEs Resembles Distance Learning and Requires Overlapping Data" ๐ This repo is an old fork of the disent framework that contains the original research code in its final state!
๐ป Script that checks if the University of the Witwatersrand's Computing Cluster is down and posts the status to Discord. The script should be deployed to AWS Lambda and triggered by EventBridge.
๐งโโ๏ธ๐โจ Magic the Gathering image dataset with automatic multithreaded downloading and optional conversion to HDF5.
Machine Learning Models for Generating MTG Cards
A fork of pyupgrade intended to selectively choose upgrades and remove some opinionated rules.
๐ต๏ธ๐ Resolve and generate requirements across modules using code scanning and the python AST.
โฑ๐ Additional resources for "Accounting for the Sequential Nature of States to Learn Features for Reinforcement Learning" submitted to RLDM 2022
๐งฌ Modularised Evolutionary Algorithms For Python with Optional JIT and Multiprocessing (Ray) support. Inspired by PyTorch Lightning
Image processing in Python
pythonflow decorator for generating dataflow graphs using natural syntax. [UNMAINTAINED]
๐๐ Python flexbox layout engine bindings for Stretch ๐
Official PyTorch implementation of StyleGAN3
๐ Tonic is a lightweight configuration framework for Python, combining notable aspects of Gin and Sacred.
A repo to host random files