Planned Enhancements

• Improved sample selection when one of the subsampling options (n or frac) is activated: sampling will consider the output distribution, whether for classification or regression.

• Automatic smoothing of output trends in apyxl.TimeSeriesNormalizer.

• Numerical experiments will be conducted to compare the capabilities of apyxl with traditional econometric techniques.

• Development of an apyxl.DiffInDiff class.

• Model saving functionality.

• Early stopping during the fitting process.

• Automatic labeling of meaningful and non-meaningful SHAP values (e.g., are their absolute values large enough? Considering using an ensemble of xgb models to discard isolated large values).