scikit-learn predict() new data

0
votes

i have a question regarding the predict() function from scikit-learn. I try to validate my trained classifier by testing with data, that is not in the training data and also has a different label. So i basically want, that the classifiers output is: 'new data is not predictable'. How do I implement that? Right now the classifier is just trying to predict the trained labels on the new data with totally different labels. Could you help me out?!

Cheers Richard

pythonmachine-learningscikit-learnpredict
The classifer doesn't look at the label when classifying/predicting new data. You should just check if the label is not in the list of labels you trained your model on - rej
Yeah, it is not in the labels of my trained model. No i do not want the classifier to look on the label. But I want the classifier to notice, when a new case is not similar at all to the trained cases, and not just trying to fit any label to it. - nollinger
because it always tries to predict any of the trained labels. It is never like, the new case is just not similar at all, so no label fits - nollinger