Distinguish fit and predict for Hierarchical-model

[ColinChrist]

Background: Since the class Hierarchical only holds meta-data and does not store anything when fit(X) is called, we can't make a prediction on new data. Ok sure, we can make a new fit, but that would also mean new clusters.

Task:

• Store metadata-information in the model about every cluster found while executing fit(X)
• Add predict-function that utilizes the cluster-information by predicting the existing clusters for new series

Please comment, if there is a better way to archive this

[wannesm]

The Hierarchical class returns a flat clustering. If the conditions for merging are loose (which is the default), there is only one cluster.

If you want to retain the tree, you can use either the HierarchicalTree or the LinkageTree class. The former is our custom algorithm, the latter uses the (faster) SciPy clustering algorithm. In both cases, after calling fit the tree is available in model.linkage.

We use linkage as defined in the SciPy package. Therefore, we do not foresee methods to operate on the linkage matrix in this package and recommend the SciPy package where this is already available.