Ole

Will close this because of other ticket "Experiment with binary trees". Will try to do that soon.

So I added a new hash called RandomBinaryProjectionTree that uses a binary tree to guarantee always N results (when combined with a NearestFilter(N)), if you still are looking for that...

I am not sure, have not read the annoy source code yet. This is kind of a naive implementation, not really tested for quality / performance. So getting benchmark feedback...

Do you have predefined configurations for the hashes? I mean the projection count? Because for given feature space dimension and data set size the projection count is relevant to the...

Ok. If you want to check different configurations with the new NearPy hash, you could vary the projection count, that is the main parameter there.

So apparently the RandomBinaryProjectionTree sucks because it is not paying respect to Hamming distance (Xing's contribution made that clear). I am currently experimenting with different approaches that pay respect to...

Sorry guys, I will try to find time for this in November.

Good point. An option to use the alpha channel as a weight in the calculation would be handy.

@xieqihui Thanks! Could you do me one more favor and add tests for the new method _store_many_vectors_ ? Sorry for not mentioning this before, but now that more people are...

@amorgun @sarthakTUM Sorry for the long silence. Last year was crazy because of Covid, work and family. I do not have much time for this project anymore. Are you interested...