This looks like a great package. Many vector-search engines do not allow you to implement your own custom distance metrics. But Unum does. Love it!
We've just hosted one of our first community/contributor calls a few hours ago, discussing the plans for the upcoming 1.0 release, and integration with UCall, UStore, and UForm - our other FOSS libraries. Please don't hesitate to reach out for any questions or feature requests - now is the best time :)
If someone here has free time and C++ experience I am open to recommendations on the codebase and style as well: https://github.com/unum-cloud/usearch/blob/main-dev/include/...
The simplest way with USearch - concatenate 10 embeddings, define a custom metric with Numba, that takes the average of 10 dot-products. Done :)
> Yes, and yes. The last one may be a bit trickier through Python bindings today, but I can easily include that in the next release… shouldn’t take more than 50 LOC.
Appreciate it. That'd be game-changing for me. The ultimate thing I'd like to do is actually use a function of the form score = af1(embedding1) + bf2(embedding2) + ...
That way you could make adjustments like ignoring feature1 unless its score passes a threshold. I'll try looking at Numba to see if that's possible.