The simplest TF-IDF library imaginable.

Add your documents as two-element lists [doc_name, [list_of_words_in_the_document]] with add_document(doc_name, list_of_words).

table.add_document("foo", ["alpha", "bravo", "charlie", "delta", "echo", "foxtrot", "golf", "hotel"])

Get a list of all the [doc_name, similarity_score] pairs relative to a list of words by calling similarities([list_of_words]). Resulting similarities will always be between 0.0 and 1.0, inclusive.

table.similarities(["alpha", "bravo", "charlie"])

So, for example:

from tfidf import TfIdf

table = TfIdf()
table.add_document("foo", ["alpha", "bravo", "charlie", "delta", "echo", "foxtrot", "golf", "hotel"])
table.add_document("bar", ["alpha", "bravo", "charlie", "india", "juliet", "kilo"])
table.add_document("baz", ["kilo", "lima", "mike", "november"])

print table.similarities(["alpha", "bravo", "charlie"]) # => [['foo', 0.6875], ['bar', 0.75], ['baz', 0.0]]

The tests use the standard library's unittest module, so there's no need to install anything. Just run:

$ python test_tfidf.py

This library is a pretty clean example of how TF-IDF operates. However, it's totally unconcerned with efficiency (it's just an exercise to brush up my Python skills), so you probably don't want to be using it in production. If you're looking for a more heavy-duty Python library to do information retrieval and topic modeling, I'd suggest taking a look at Gensim.