Fix SimilarityEncoder n-gram counts past the end of the string

[SashaMIT]

SimilarityEncoder(ngram_range=(2, 5)) fit on "a" returned self-similarity 3. get_ngram_count adds len(string) - n + 1 for every order, including windows longer than the string, so the padded string " a " (length 3) was counted as 2 n-grams instead of 3. The fast path, which is the default, then divided the overlap by that short total. The slow path uses the vectorizer totals and returned 1. Orders that do not fit now contribute 0. The default ngram_range=(2, 4) on "a" is unchanged, because the order-4 window count was already 0.

test_similarity_encoder_self_similarity_when_ngram_exceeds_string failed on main (3) and passes after the change. test_similarity_encoder.py passed (13 passed, 9 xpassed). The rest of the package pytest suite was not run.

AI Disclosure

• This PR contains AI-generated code
• I have tested the code generated in my PR
• I have read and understood every line that has been generated by the AI agent
• I can explain what the AI-generated code does

[rcap107]

Hi @SashaMIT, out of curiosity, did you fix this because you encountered the issue while using the SimilarityEncoder?

[SashaMIT]

Hi, I did not hit this while using SimilarityEncoder. I was reading the n-gram counts, and a window longer than the string came out negative, so the fast path divided by 2 instead of 3. The slow path on the same input was 1.

[rcap107]

Looks good to me, thanks @SashaMIT