Add svr benchmark - Githubissues

mlpack / benchmarks

Machine Learning Benchmark Scripts

101 stars 49 forks source link

Add svr benchmark #11

Closed Saurabh7 closed 8 years ago

Saurabh7 commented 8 years ago

@karlnapf Porting one the benchmarks , Support vector regression Results:

                   scikit      shogun 
diabetes         0.017914    0.024873 
cosExp           0.005586    0.002809 
TomsHardware   128.226224  133.479343

vigsterkr commented 8 years ago

@Saurabh7 thnx a lot! :) maybe some of the functions in SVR should be reviewed :P

Saurabh7 commented 8 years ago

@vigsterkr looks like it ;)

karlnapf commented 8 years ago

Great, thanks! Which SVM solver is used here? Is it the same as for sklearn? If no, make sure it is. Also make sure the parameters (convergence epsilon etc) are the same, as well as the kernel used.

karlnapf commented 8 years ago

It would be cool to also measure time needed for model selection of kernel parameter and C. Will put up an issue

karlnapf commented 8 years ago

https://github.com/shogun-toolbox/shogun/issues/3114

Saurabh7 commented 8 years ago

Yes both use the epsilon version with same parameters and also gaussian kernel for both ..

karlnapf commented 8 years ago

Find to merge from my side!

zoq commented 8 years ago

Looks good to me; merged in 63d4a7bdff0486e1. Thanks for the contribution.