Install dependencies locally, dev-install the package, download data (only needs to be done once).
Get the dataset on Kaggle.
# from repo root
poetry install
# requires kaggle creds
mkdir -p data; cd data
poetry run kaggle competitions download -c house-prices-advanced-regression-techniques
unzip house-prices-advanced-regression-techniques.zip
cp transforms data/
Running tests
poetry run pytest
Interactive jupyter notebooks
poetry run jupyter notebook
Training
poetry run python -m house_prices_mlp.main
- mix out or mix in data augmentation
- cosine annealing for the learning rate
- clu logging
Less important / more ambitious
- ml collections params
- lifted module with jit?