lesshaste

The example I showed was with N reduced to 7 so that it runs with symfit 0.51. I haven't tried symfit master with the new fix included which is what...

That's a very nice table, thank you. It's a shame that with the breadth of optimizers only SLSQP seems to work at all for my simple problem. Out of interest,...

Thank you for this. So if I understand correctly, TrustConstr is in fact the only optimizer that uses the Hessian and this will still fail with the "too many parameters"...

This looks great and solves an important problem. It would be wonderful if it could be reviewed and merged.

Doesn't your example show hnsw to be *much* slower than brute_force? That is 102 seconds versus 8383 seconds? I also ran your code on my PC and brute_force takes about...

Sorry I meant something a little different. I changed your code to query with the whole of seqs (not seqs[0:queryQty]). The full output is now: ``` Made data. Now time...

That sounds very interesting!

I don't really make a distinction but let me be more precise. I will write HD for Hamming distance. HD('0215', 0211') = 1 HD([0,2,1,5], [0,2,1,1]) = 1 So the difference...

Right, I don't want L1 in this case. It's true that for strings it's hard to get an alphabet larger than 95. However, nmslib could accept also vectors of integers...

I do feel it would be a great addition. I can think of quite a lot of uses myself.