shogun-toolbox / shogun

Shōgun
http://shogun-toolbox.org
BSD 3-Clause "New" or "Revised" License
3.03k stars 1.04k forks source link

add cookbook examples for GPs #3091

Open karlnapf opened 8 years ago

karlnapf commented 8 years ago

We have a lot of different inference algorithms and likelihoods for GPs in shogun. This task is to write a cookbook page for each of them. Many entrance tasks in fact. Please state which one you are working on here to avoid redundant work :)

TianqiTang commented 8 years ago

New comer here. Sorry but have to ask what is GP short for? Gaussian Process?

karlnapf commented 8 years ago

Yes

On Saturday, 19 March 2016, Tianqi Tang notifications@github.com wrote:

New comer here. Sorry but have to ask what is GP short for? Gaussian Process?

— You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub https://github.com/shogun-toolbox/shogun/issues/3091#issuecomment-198746804

ialong commented 8 years ago

@karlnapf Hey Heiko, I wanted to get started on this. If it's ok with you I thought I would start with the three available likelihood functions for GPC (LogitVGLikelihood, LogitVGPiecewiseBoundLikelihood, LogitDVGLikelihood) and then move on to the GPC inference methods. Let me know if there's something else you'd like me to cover first

ialong commented 8 years ago

also just an aside note, I ran both GP ipython notebooks and both seem to crash at some point to a "GradientModelSelection" not being defined. Is there something deep/worrisome going on or has just the syntax changed to a new name?

karlnapf commented 8 years ago

This seems to be a name change. Feel free to send a fix for those notebooks... As for GP, I think the best would be to have a sperate example for all use cases that we have

Start with standard GP regression, standard GP classification. Keep the notebooks simple :) BTW I think we can have a seperate gp folder in the cookbook...

yorkerlin commented 8 years ago

@Ialong Let me know whether there is any issue related to GP. If you can fix these issues, please send a PR for each issue.

ialong commented 8 years ago

@yorkerlin I have looked at the GradientModelSelection class and it seems like it's still there and it looks fine. For some reason it cannot be found within the python module. GridSearchModelSelection and the other related classes seem to be fine. Do you have any ideas as to what might be causing this?

karlnapf commented 8 years ago

check the code #ifdef HAVE_NLOPT This is what you need

yorkerlin commented 8 years ago

@Ialong which notebook? http://shogun-toolbox.org/static/notebook/current/gaussian_processes.html or http://www.shogun-toolbox.org/static/notebook/current/variational_classifier.html

yorkerlin commented 8 years ago

@Ialong Shogun needs NLOPT lib for GradientModelSelection. Make sure that the lib is installed in your system.

ialong commented 8 years ago

@karlnapf @yorkerlin Ah of course, thank you! It works now

ialong commented 8 years ago

@karlnapf A GP regression cookbook is ready. It's very basic as I did not want to make it too long. Should I extend it to include things like sparse approximations or should I make a separate notebook for that?

Also I could not find information on the cookbook README page about adding new cookbooks to the general cookbook index. Is there something I need to do or should it happen automatically?

karlnapf commented 8 years ago

Hi, the cookbook pages should be very simple and minimal. Rather add new pages for sparse inference methods.

the readme doesnt cover changing the index. Feel free to send an update with a mini explanation

karlnapf commented 8 years ago

3054 the notebook in here has the way to learn things with ml2