LAMA, or LightAutoML is open-source python library released by the AutoML Team at Sber AI Lab as an Automated Machine Learning (AutoML) framework. It is designed to be lightweight and efficient for various tasks:
- binary/multiclass classification,
Now, LightAutoML supports tabular datasets, which contain different types of features: numeric, categorical, dates, texts, images, etc.
LightAutoML installation is pretty simple:
pip install -U lightautoml
LightAutoML provides not only presets for end-to-end ML task solving but also the easy-to-use ML pipeline creation constructor, including:
- data preprocessing elements,
- advanced feature generation,
- CV schemes (including nested CVs),
- hyperparameters tuning,
- different models, and composition building methods,
- model training and profiling reports to check model results and find insights that are not obvious from the initial dataset.
Currently, LAMA supports datasets, where each row is an object with its specific features and target. Multitable datasets and sequences are now in the development phase.
Note: LAMA has automatic creation of interpretable models onboard that is based on AutoWoE library made by our team as well. Interpretable blackbox is on the way to release!
Docs, tutorials and main repository: GitHub, AutoWOE
LightAutoML Communities: Telegram Chat, Slack
Team: Maksim Savchenko, Alexander Ryzhkov (telegram, slack: @RyzhkovAlex), Anton Vakhrushev (slack: @btbpanda), Dmitry Simakov, Vasilii Bunakov, Rinchin Damdinov, Pavel Shvets, Alexander Kirilin,