Implementation of RuLSIF

hoxo-m / densratio_py

A Python Package for Density Ratio Estimation

https://github.com/hoxo-m/densratio_py

Other

127 stars 30 forks source link

Implementation of RuLSIF #4

Closed ameya98 closed 5 years ago

ameya98 commented 5 years ago

Computes the alpha-relative density-ratio via RuLSIF, a modification to uLSIF. (When alpha is 0, these are identical). Further, contains an implementation of the alpha-relative KL-divergence and alpha-relative PE-divergence approximators for the computed density-ratio approximator.

Reference: Change-Point Detection in Time-Series Data by Relative Density-Ratio Estimation Song Liu, Makoto Yamada, Nigel Collier and Masashi Sugiyama https://arxiv.org/pdf/1203.0453.pdf

ameya98 commented 5 years ago

Added some more tests as well - plots in test_advanced.

ameya98 commented 5 years ago

Also, I fixed the issue with sigma_range and lambda_range defaults. Previously, they did not span from 1e-3 to 1e9, while the function docstring claimed they did.

coveralls commented 5 years ago

Coverage increased (+0.7%) to 96.38% when pulling 7134332165cd7138d34abc498a62e87cc2f0c699 on ameya98:master into dd9106c5db45112b8047823588da29d38fa54bb0 on hoxo-m:master.

hoxo-m commented 5 years ago

It is a great job!

hoxo-m commented 5 years ago

I think you should be added as an author because you made a big contribution. Could I do it?

ameya98 commented 5 years ago

Sure, that would be amazing! I'm Ameya Daigavane. Thank you so much!

ameya98 commented 5 years ago

The documentation also needs to be changed slightly to reflect these changes. I will get to that soon.

ameya98 commented 5 years ago

Documentation changes are done, see my other PR. Thanks!