shogun-toolbox / shogun

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

kernel comparision in Shogun #3135

Open yorkerlin opened 8 years ago

yorkerlin commented 8 years ago

@karlnapf reference: http://scikit-learn.org/dev/auto_examples/classification/plot_classifier_comparison.html

yorkerlin commented 8 years ago

For example, we can use different kernels and inference methods.

yorkerlin commented 8 years ago

@karlnapf I think it will be great if we can automatically include examples/cookbooks in Shoung's API doc.

See Examples using sklearn.gaussian_process.GaussianProcessRegressor in sklearn's GPR

http://scikit-learn.org/dev/modules/generated/sklearn.gaussian_process.GaussianProcessRegressor.html#sklearn.gaussian_process.GaussianProcessRegressor

karlnapf commented 8 years ago

Integrating this is exactly the point of the cookbook. So yes, once GSoC is over, we have many more than now and we can do this.

BTW there already is a notebook that compares the classifiers #3019

yorkerlin commented 8 years ago

@karlnapf this issue is for kernel machine only :). For classification in GP, there are many components to choose. eg, kernels, likelihoods (eg, logistic, probit), inference methods.

karlnapf commented 8 years ago

Why would that be different from what we have?

yorkerlin commented 8 years ago

Decision boundary may be different even when training accuracy is the same since we use different kernels/likelihoods/inference methods.

yorkerlin commented 8 years ago

@karlnapf for example, Decision boundary in linear SVM is different from the one in RBF SVM.

reference: http://scikit-learn.org/dev/auto_examples/classification/plot_classifier_comparison.html

karlnapf commented 8 years ago

Maybe we are misunderstanding each other: Such plots are already in that notebook I referenced above. Kernel SVM vs linear SVM vs GP. I don't see a reason to do another comparison for GPs. Why do you wanna do that?

yorkerlin commented 8 years ago

For example, GP using RBF kernel vs GP using linear kernel.

reference: http://people.seas.harvard.edu/~dduvenaud/cookbook/

yorkerlin commented 8 years ago

BTW, only GP with RBF kernel is used in that plots.

karlnapf commented 8 years ago

Well I think this is independent of GPs. What about a cookbook on covariance functions and kernels in Shogun?

yorkerlin commented 8 years ago

https://github.com/yorkerlin/plots-for-GP/blob/master/Untitled1.ipynb

karlnapf commented 8 years ago

Do you maybe want to put that in a new (polished) version and put in a separate notebook?

yorkerlin commented 8 years ago

Add a notebook/cookbook about http://people.seas.harvard.edu/~dduvenaud/cookbook/ using the Shogun's GP.