https://github.com/patrickfrank1/chess-embedding-experiments
Experimentation
https://github.com/patrickfrank1/chess-embedding-experiments
Last synced: about 1 year ago
JSON representation
Experimentation
- Host: GitHub
- URL: https://github.com/patrickfrank1/chess-embedding-experiments
- Owner: patrickfrank1
- Created: 2023-08-02T19:07:51.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2024-04-01T11:45:40.000Z (about 2 years ago)
- Last Synced: 2025-02-08T20:12:51.734Z (over 1 year ago)
- Language: Jupyter Notebook
- Size: 1.03 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Chess Embedding Experiments
## Cheat Sheet
- Generate training positions
python -m src.run.generate_positions
- Train a neural network
python -m src.run.train
- Evaluate a trained network by starting the notebook: `src/run/evaluate.ipynb`
## Tools
- Start ML Flow UI, in correct python venv
mlflow ui
- Export dependencies to requirements.txt
poetry export > requirements.txt
## Notes
### Milvus
- could only get milvus 2.3.1 to work, so use that for now
- but had to downgrade python to 3.9, because of compatibility issues
- and only works with recent tensorflow version, so it's incompatible with aws sage maker
- maybe I need to build a different toolchain for different python versions
## TODOs
- [ ] Document findings of up to current model training
- [ ] Write to db from .npy files
- [ ] write tokenized positions with some metadata and id
- [ ] write embeddings generated from a model
- [ ] Write to db from .pgn file
- maybe some refactoring is needed
- make embeddings better for search
- document approaches
- make a plan