Preprocessing that Classify[ ] does on NumericalVectorSequence[ ]?

Questions
,
1

When the input data for Classify is a list of vector sequences where the number of vectors in each sequence is not constant, called NumericalVectorSequence, Classify performs a preprocessing step called NumericalSequenceExtractFeatures that converts NumericalVectorSequence to NumericalVector. I was wondering what this step does. When I look inside the returned ClassiferFunction and hover over NumericalSequenceExtractFeatures, I see that some Chi-squared test is being done, but it is not exactly clear what Mathematica is doing under the hood to preprocess a list of vector sequences. I attached a relevant notebook with some sample NumericalVectorSequence type data and a ClassiferFunction that does the preprocessing step that I am concerned with.

vector_seq.nb (1.03 MB)

2

+1 It would be great to have documentation about these things.

These super-powerful functions (Classify and Predict) are really difficult to use in production environments as long as the documentation of procedures employed is so sparse and the procedures themselves are impossible to modify :frowning:

There are clearly a lot of options for how to preprocess general data. It would be super helpful if these were exposed to the user. At the very least, it would be cool if we could see what assumptions and procedures were done (and have associated high-quality documentation instead of having to reverse-engineer procedures from vague parameter names in <|Associations|>)