I have already extracted image features of 3 classes using SURF (openCV) and stacked all descriptors together into a numpy array.
My next steps would be:
- clustering using K-means
- build feature histogram for each training case based on clusters
- train a multi class linear SVM to separate cases based on feature histograms
My question is: at what point would it be most appropriate to scale my data?
Should this be done before clustering, histograms or SVM?
I am using sklearn's standard scaler, as well as linearSVC.