Sorry it was not 'scoring' but 'pred' in your first example.
In your current example, you should take the opposite of the OCSVM decision
function.
(by convention, bigger is better (here more normal), contrary to your
labels)
Nicolas
2015-10-23 3:56 GMT+02:00 Ady Wahyudi Paundu <awpau...@gmail.com>:
> Hi Nicolas, Thank you for the pointer.
>
> Since i didn’t know where to take the value of ‘scoring’ variable,
> this is what I did (by adapting to the example in scikit-learn web)
>
> …
> X_test = np.r_[ND, AD]
>
> # fit the model
> clf = svm.OneClassSVM(nu=float(NU), kernel="rbf", gamma=float(GA))
> scoring = clf.fit(LD).decision_function(X_test)
>
> pred = clf.predict(X_test)
> y_true = np.array([0]*500+[1]*500)
> fpr,tpr,thresholds = roc_curve(y_true,scoring)
> roc_auc = auc(fpr,tpr)
>
> plt.figure()
> plt.plot(fpr, tpr, label='ROC curve (area = %0.2f)' % roc_auc)
> plt.plot([0, 1], [0, 1], 'k--')
> plt.xlim([0.0, 1.0])
> plt.ylim([0.0, 1.05])
> plt.xlabel('False Positive Rate')
> plt.ylabel('True Positive Rate')
> plt.title('Receiver operating characteristic')
> plt.legend(loc="lower right")
> plt.show()
>
> …
>
> this however showed only the axis and diagonal dots without ROC curve
> (auc=0).
> How to fix this?
>
> Best,
> Ady
>
> On 10/20/15, Nicolas Goix <goix.nico...@gmail.com> wrote:
> > Hi,
> > You can do:
> >
> > ```
> > from sklearn.metrics import roc_curve, auc
> >
> > X_test = np.r_[ND, AD]
> > pred = clf.predict(X_test)
> >
> > y_true = np.array([0] * 500 + [1] * 500)
> >
> > fpr, tpr, thresholds = roc_curve(y_true, scoring)
> >
> > # then you can plot(fpr, tpr) to get the roc curve and compute the AUC
> > with:
> > AUC = auc(fpr, tpr)
> >
> > ```
> >
> > Best,
> > Nicolas
> >
> > 2015-10-20 3:41 GMT+02:00 Ady Wahyudi Paundu <awpau...@gmail.com>:
> >
> >> Hi all,
> >>
> >> Can I create ROC curve for one_class_SVM classifier?
> >> If I can, can you give pointer on how to do this? (or a link?)
> >>
> >> for example now i have:
> >> LD: normal data for learning (100 item)
> >> ND: normal data for evaluation (500 item)
> >> AD: abnormal data for evaluation (500 item)
> >>
> >> one_class_SVM code for evaluation would be something like this (NU and
> >> GA is user input):
> >>
> >> clf = svm.OneClassSVM(nu=float(NU), kernel="rbf", gamma=float(GA))
> >> clf.fit(LD)
> >> y_pred_train = clf.predict(LD)
> >> y_pred_test = clf.predict(ND)
> >> y_pred_outliers = clf.predict(AD)
> >> n_error_train = y_pred_train[y_pred_train == -1].size
> >> n_error_test = y_pred_test[y_pred_test == -1].size
> >> n_error_outliers = y_pred_outliers[y_pred_outliers == 1].size
> >>
> >> how to produce ROC curve from there?
> >>
> >> Thanks in advance
> >> ~Ady
> >>
> >>
> >>
> ------------------------------------------------------------------------------
> >> _______________________________________________
> >> Scikit-learn-general mailing list
> >> Scikit-learn-general@lists.sourceforge.net
> >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
> >>
> >
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-general@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
------------------------------------------------------------------------------
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
