Skip to content

AutoML-AutoDL

AutoML-AutoDL is a tabular risk modeling workspace for the Home Credit Default Risk style binary classification task. The input file is data/application_train.csv, and the target column is TARGET.

The project has one shared preprocessing pipeline and two modeling tracks:

  • AutoML: classical machine learning models trained with a shared GridSearchCV wrapper.
  • AutoDL: PyTorch and TabNet based deep tabular baselines trained on the same feature matrix.

Quick run:

pip install -r requirements.txt
python main.py
python autodl/train.py

AutoML writes:

outputs/automl_grid_search_results.json

AutoDL writes:

outputs/autodl_training_results.json

What is covered

  • Quick start: dependencies, AutoML, AutoDL, Docker, and docs commands.
  • Data and preprocessing: categorical columns, numeric columns, missing values, one-hot encoding, scaling, and passthrough columns.
  • AutoML: shared grid search, metrics, refit strategy, and 13 model-specific grids.
  • AutoDL: shared training loop, validation split, early stopping, and 5 model structures.
  • Project notes: device selection, output files, repository layout, and PRDs.