v1.6.4

This release adds more robustness and increases performance a bit.

• target encoding will be executed in cross-validated fashion instead of just simple target encoding outside of folds
• datetime features will also produce cyclic transformations of those. To prevent overfitting we ignore all date components that have just one unique value.
• calibrate methods will automatically convert to pandas series if a numpy array has been provided
• Domain checker and Modelmatchmaker support GPU acceleration now
• hyperparameter tuning will not tune number of estimators by default. It is fixed to 1000, but can be customized