The number of points to sample should decrease as we move forward with the classification.
As we move forward with the classification, the number of points to classify should reduce. Hence, the increase of init.samples
should not lead to a big performance degradation.