LionelMassoulard
LionelMassoulard
WIP Documentation on some advanced functionnalities of the GraphPipeline
https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.HistGradientBoostingClassifier.html
In the "fit" command of the automl, allow the parameter "best" to directly fit the best model (instead of specifying the model to fit by its job_id)
Leverage existing tests within aikit to offer a toolbox to check behavior of new klass. The toolkit would : - test what data structured is accepted and not (DataFrame, numpy,...
Idee : facilite the creation of a model that actually fit one model per type of modality. Example : fit on model per country if a country. The code could...
Allow the auto-ml to use lgbm advanced capabilities: - internal cv to use early stopping - automatic gestion of categorical variable
Make shap explanation work will with a pipeline
add things to handle temporal series prediction - use a temporal cv - create special transformer (looking back to create feature for exemple)
Add typing indication to most important function