Changed - Change parameter order in `hcl_model.model_base.ModelBase.fit` and `.predict`. This is important if the estimator is used in `sklearn.pipeline.Pipeline` object. - Allow `np.ndarray` for `y` input of the `fit` method. In this case its index is inherited from `X`. If `X` is `None`, then `TypeError` is risen.
0.5.2
Changed - Store all transformer parameters as class attributes with the same name during initialization.
0.5.1
Changed - Convert previously private method to public: `hcl_model.model_base.ModelBase.get_parameters`.
0.5.0
Added - `sklearn` compliant transformer `TargetOutlierCorrectionTransformer` in `hcl_model.transformers.outlier_correction`. Read documentation in `docs/data_preprocessing.md`. - `sklearn` compliant transformer `TargetStructuralBreakCorrectionTransformer` in `hcl_model.transformers.structural_breaks`. It uses [`ruptures` package](https://github.com/deepcharles/ruptures/) to correct structural breaks in `X` data. - `sklearn` compliant transformer `EstimatorToTransformer` in `hcl_model.transformers.estimator_to_transformer`. Effectively, it replaces `predict` method with `transform` so that the object is usable as an intermediate step in `Pipeline`. - `sklearn` compliant transformer `TruncateTransformer` in `hcl_model.transformers.truncate`. Leave only a specific number of past weeks in the data. - `sklearn` compliant transformer `SeriesToFrameTransformer` in `hcl_model.transformers.series_to_frame`. It simply applies `.to_frame()` to a `pd.Series`. - A replacement for `sklearn.pipeline.FeatureUnion`: `DatetimeIndexedFeatureUnion` in `hcl_model.transformers.feature_union`. - A replacement for `sklearn.pipeline.make_union`: `make_union_of_datetime_indexed_features` in `hcl_model.transformers.feature_union`. - Move utility functions from `utils.py` to a separate module `utils` with `.py` file per function. Changed - Move `CalendarTransformer` to `hcl_model.transformers.calendar`. - Move `CalendarReader` to `hcl_model.utils.calendar_reader.CalendarReader`.
0.4.0
- Rename arguments: `exog` into `X`, and `endog` into `y`. This is done to comply with [`sklearn`](https://scikit-learn.org/stable/developers/develop.html) general interface. - Now `.fit` should be fed only with past data, and `.predict` only with future data.
0.3.9
- Allow passing arbitrary transformations to `HandCraftedLinearModel` that return multi-column dataframes.