18. EC2021: Automated Machine Learning for Earth Science via AutoGluon


Notebook that demonstrates the usage of AutoGluon at EarthCube 2021.

You may learn how to use AutoGluon via the notebook in Notebook.

To train a model for Forest Cover Type Prediction with the best_quality option, you can try the following command.

wget https://deep-earth.s3.amazonaws.com/datasets/earthcube2021_demo/forest-cover-type-prediction.zip -O forest-cover-type-prediction.zip
unzip -o forest-cover-type-prediction.zip -d forest-cover-type-prediction
python3 train_forest_cover_type.py

This will generate a file named stack_ensemble_submission.csv and a model that is saved to forest_cover_ensemble. We used the same script to reach 82/1692 in the competition. The screenshot for our previous run is provided in the following figure.


Feel free to checkout our website: https://auto.gluon.ai/