The random forest embedding needs modified distance metrics

[ShumingXu]

Description

The current random trees embedding in Sklearn gives the leaf node that samples land into as the output. It is only used as a way to transform the dataset to a higher dimension in the given examples. To make it a better algorithm for unsupervised clustering and classification, I plan to introduce three distance metrics that can help boost the classification performance.

Planned Enhancement in the Form of PR

• Implement the algorithm to generate three different distance metrics: 1) depth of nearest common ancestor; 2) length of shortest path; 3) proximity matrix from random trees embedding estimators using scikit-learn package
• Give examples on how to choose clustering algorithm and parameters to be used on the output of RandomTreesEmbedding

[bdpedigo]

You need a lot more detail about the things you are going to implement. And what it is you will actually be PRing.