Sklearn-pmml-model

Bugfixes
- Fixed a problem that could block logistic regression from working in new versions of scikit-learn (4b6e11cf5f96151eb9fa49e58c3a89ee1c870ca2).
- The input array shape now gets validated, and pandas dataframes gets subscripted and reordered to match the PMML file (2555898d3b3853eb100a4025de9bd98b091a7074).

Enhancements
- 🌟Added support for support vector machine classification and regression (31).
- Several improvements to documentation and code style.

Bugfixes
- Fixed an issue with categorical features that occurred when categories contained spaces.

Enhancements
- 🌟Added support for tree-based regression, including decision trees (`PMMLTreeRegressor`), random forests (`PMMLForestRegressor`) and gradient boosting (`PMMLGradientBoostingRegressor`) (25).
- Added support for classification with linear models, with `PMMLLogisticRegression` for regression models, and `PMMLRidgeClassifier` for general regression models (24).

Bugfixes
- Fix compatibility issues with the latest version of scikit-learn
- prevent inheritance problem related to `_more_tags` property

Enhancements
- 🌟Support gradient boosting classifiers, including categorical features (20)
- Add an example that highlights the usage of `tree.decision_path`

Bugfixes
- Correctly parse ensemble trees with only a subset of target features. This happened to be the case for certain PMML models created with MatLab (21)

Bugfixes
- Prevent Buffer dtype mismatch on 64 bit windows (reported in 19)

Bugfixes
- LocalTransformations are now parsed rather than only the global TransformationDictionary

Enhancements
- Added support for multiSplit decision trees
- Support average multipleModelStrategy for Random Forests