Quick start¶

Install dependencies¶

Run this from the repository root:

pip install -r requirements.txt

The training environment includes pandas, numpy, scikit-learn, torch, pytorch-tabnet, xgboost, lightgbm, and catboost. If an optional library is missing, the matching model returns a structured skipped result instead of stopping every other model.

The documentation site has its own dependencies in docs-site/requirements.txt.

Run AutoML¶

python main.py

This command:

Loads and preprocesses data/application_train.csv.
Runs the models listed in config.AUTOML_MODELS.
Saves the summary to outputs/automl_grid_search_results.json.

Run AutoDL¶

python autodl/train.py

This command runs MLP, BLSTM, CNN1D, TabNet, and Transformer, then writes:

outputs/autodl_training_results.json

Docker¶

Build:

docker build -t automl-autodl .

Run AutoML:

docker run --rm automl-autodl

Run AutoDL:

docker run --rm automl-autodl python autodl/train.py

The Docker path mainly covers CUDA and CPU. MPS detection remains in the code for native Apple Silicon PyTorch runs.

Documentation site¶

cd docs-site
pip install -r requirements.txt
mkdocs serve

Strict build:

mkdocs build --strict