Version 3 not in pypi repo

Not sure if this was the right place, but I want to upgrade my imagehash version and doing pip install imagehash --upgrade still gives me the 2.2 version, presumably because the pypi page hasn’t been updated with v3.

[JohannesBuchner]

numpy.flatten makes a 1d array out of the 2d array (matrix) where the hash is stored. != compares element-wise if the things are the same, numpy.sum() adds the number of differences, which gives the Hamming distance.

What I am suggesting is:

I suspect you are doing something like this now:

diff, closestdbHash = min([(dbHash - testhash, dbHash) for dbHash in db])

which does the same as

diff, closestdbHash = min([((dbHash.hash != testhash.hash).sum(), dbHash) for dbHash in db])

You can store your DB as a 2d matrix:

arr = []
for dbHash in db:
    arr.append(dbHash.hash.flatten())
arr = numpy.array(arr)

Then you can do the comparison against all at the same time:

binarydiff = arr != testhash.hash.reshape((1,-1))
hammingdiff = binarydiff.sum(axis=1)
closestdbHash_i = numpy.argmin(hammingdiff)
closestdbHash = db[closestdbHash_i]

This will be very, very fast.

[mynameisvinn]

@PAK90 @JohannesBuchner ive been using approximate nearest neighbor to loop through a collection of hashes. annoy was developed by spotify to find similar users/songs for their recommender system.

https://github.com/spotify/annoy

hope this helps!