search

SheffieldML / GPy

Gaussian processes framework in python
BSD 3-Clause "New" or "Revised" License
2.04k stars 562 forks source link

New heteroscedastic GP model #619

Open col14m opened 6 years ago

col14m commented 6 years ago

Hello As part of my scientific work at the institute, I need to implement a new heteroscedastic model from the article "Variational Heteroscedastic Gaussian Process Regression" in GPy. I would like to know if this is an actual topic and how interesting it will be for you. As far as I know, there was an attempt to add this model to your library, but it failed, because several errors were made. If you are still interested, then I can follow your instructions correctly to implement this model.

mzwiessele commented 6 years ago

We're always looking to expand GPy and appreciate any pull requests :+1:

col14m commented 6 years ago

It's great! First of all, how much I understand the main error in this code (https://github.com/SheffieldML/GPy/pull/307) was the use of its own kernel. Secondly, it was necessary to inherit from the GP class, but not from the Model.Also there were errors associated with the calculation of gradients. I do not quite understand how the optimization process works on parameters. There is very little in the documentation about this. Although this is probably the most basic problem when creating a model. Also, this code does not work on the latest version of the library, because many modules related to optimization parameters have been changed. I would really like to understand how this works.