scikit-learn-contrib / hiclass

A python library for hierarchical classification compatible with scikit-learn
BSD 3-Clause "New" or "Revised" License
114 stars 20 forks source link

What is the training policy on LCPPN? #61

Closed guipaiva closed 2 years ago

guipaiva commented 2 years ago

Hello,

first of all, good job on the project, I really like it!

I have seen that local classifier per node has specific training policies which define the positive and negative set of training examples. But as far as i saw, it does not mention what is the policy for local classifier per parent node.

Following this example from the docs, does the mammal classifier takes in account all data for training or just the data from mammal and it's successors classes?

Thanks

mirand863 commented 2 years ago

Hello @guipaiva,

Thank you for the interest in HiClass!

The local classifier per parent node did not need a policy implemented, since it is a multi-class classifier. However, if I were to speak in terms of policies, I guess I could say the closest policy would be the siblings policy, but with a slight modification. I only get the data regarding children nodes when training a parent node, hence, the policy would be Tr+(cj) = the set of descendant categories of cj.

I hope that answers your question. :)

guipaiva commented 2 years ago

It does answer my question. Thanks a lot!

Best,