In the cos_similarity and Centroid.normalize a divide by zero is prevented by setting a minimal vector norm of some small epsilon. When a norm less than this epsilon occurs a warning should be printed that reports this to the user because this most likely means that there is a bug. E.g., one class has no training samples.
In the
cos_similarity
andCentroid.normalize
a divide by zero is prevented by setting a minimal vector norm of some small epsilon. When a norm less than this epsilon occurs a warning should be printed that reports this to the user because this most likely means that there is a bug. E.g., one class has no training samples.