TF-IDF Vectorizer

Counting words treats "the" and "thrombosis" as equally informative. TF-IDF fixes that by multiplying how often a term appears in a document by how rare it is across the collection.

Use exactly these definitions:

1tf(t, d)  = count of t in d / number of tokens in d
2idf(t)    = ln(N / df(t))            N = number of documents
3                                     df(t) = documents containing t
4value     = tf(t, d) * idf(t)

Task: write tfidf(documents), where each document is a list of tokens. Return one vector per document, rounded to 4 decimal places.

• The vocabulary is every distinct token across all documents, sorted alphabetically. Every vector uses that same order.
• Natural log. No smoothing, no +1 anywhere, and no vector normalization — stick to the formulas above.

A term appearing in every document gets ln(N/N) = 0 and its whole column vanishes. That's the mechanism doing the real work: stopwords aren't removed by a list, they're mathematically zeroed out for being everywhere.