russelltsuchida / klr Goto Github PK

hi Russell

I just sent a pull request for #1 , #2 , and #3

I wasn't sure how we could test these functions though. Can we arrange a meeting some time next week if that's okay with you? 😇

thanks

We should allow users the option of specifying pre-computed kernel matrices instead of a kernel function and some data. For this to happen, we should check in the init method whether the passed object is a function or a kernel matrix. If it is a matrix, we should go into "matrix mode", in which case the predict and other methods should accept another kernel matrix instead of some new data x.

We should implement similar methods to those in Sklearn's logistic regression class More specifically, we should implement the following methods:

• predict p
• redict_proba
• predict_log_proba
• decision_function

This requires doing some simple manipulations to the output on line 48 of klr.py.

For a C class problem, make a container class that instantiates C klr objects and performs multi-class classification according to the Sklearn API.