Hi,
I just started to figure out how SVM for 'regression' is implemented in
PyMVPA/libsvm. In comparison to SVM for 'classification' I am not
interested in the classification accuracy but in the explained variance.
Setting up cross-validation for 'classification' is easy like:
clf = LinearCSVMC()
splitter = NFoldSplitter(cvtype=1)
cv = CrossValidatedTransferError(
TransferError(clf),
splitter)
However, here is my code for 'regression':
clf = SVM(kernel_type='sigmoid', svm_impl='EPSILON_SVR', regression=True)
dets=[]
for nfold, (training_ds, validation_ds) in enumerate(splitter(dataset)):
clf.train(training_ds)
prediction = clf.predict(validation_ds.samples)
r = corrcoef(prediction, validation_ds.labels)
det_coef = r[1][0]**2
dets.append(det_coef)
mean_det_coef=mean(array(dets))
I was wondering if there is smth equivalent to
"CrossValidatedTransferError" for regression?
Thanks,
Matthias
_______________________________________________
Pkg-ExpPsy-PyMVPA mailing list
[email protected]
http://lists.alioth.debian.org/mailman/listinfo/pkg-exppsy-pymvpa
