Dear All,
I'm stuck with a problem and I don't know if it's a bug. I'm defining the
optimization parameter C and gamma for my SVM in this way:
C = 10.0 ** numpy.arange(-3, 9)gamma = 10.0 ** numpy.arange(-6, 4)param_grid =
dict(gamma=gamma, C=C)svr = svm.SVC(kernel='rbf')clfopt =
grid_search.GridSearchCV(svr,param_grid)clfopt.fit(X_train, Y_train)
Changing the dataset I always get the same C and gamma that are 0.001 and
1e-06. With that I get a worst result!!!If I set manually to different C and
gamma I get a better result!
With this optimization I get always the same result also for different subset
of training of my dataset!
I'm completely lost and I don't know if probably this is due to the training
set size. I'm saying this because if I reduce my dataset, this optimization
works. So I was thinking that the cause could be the big dataset.
Maybe is a bug that with big dataset it give these strange results of C and
Gamma (that are the lower limit that I set for the two to be researched).
Do you know if there is another way to find a best C and Gamma without using
the grid_search_GridSearch ?
If you think strange I will segnalate as bug.
Thanks All!!!
------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:
Build for Windows Store.
http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
