This is a javascript library to perform SRP locality-sensitive hashing in browser-based situations.
The results may differ from the python version in some parts of the Unicode character space because of tokenization or lowercasing rules. The python version should be considered the reference implementation. In practice, the two versions seem to be close enough that models trained on python versions run well on Javascript versions.
The python library is much more highly optimized than the JS one.
It exposes a single function, SRP, that instantiates an object used for hashing.
Create an object which is used to hash further strings.
Object orientation is used, rather than functions, in order to build up a cache that allows faster operation.
hasher = SRP(640)Running a hash of a string
hasher.SRP("foo bar")Running a hash from word counts is also possible. The words will be re-tokenized to ensure they work with the SRP tokenization definition. The following command is equivalent to the one above.
hasher.SRP(["foo", "bar"], [1, 1])If for some reason you need the tokenization the SRP creates, you can extract it as an ordered list.
hasher.tokenize("foo bar")Poke around the code and you'll see the option to change some things including the log count settings. I don't recommend doing this.