You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I just wanted to try out this package. I've used NearestNeigbors.jl before, but need to be able to add and remove points from the tree efficiently.
I want to start with an empty index (nvectors=0) and build it up, but got stuck on this assert:
the index relies on a coarse quantizer (often just cluster ceters) and a residual quantizer, both learned from data
a simple method to build an empty index is to build it and remove all indexed points. The coarse and residual quantizers remain in the object while the points are deleted.
there is not straightforward way of incrementally building this type of index and it was never supposed to. The quality depends on the initial indexed data.
Yep, seems not. A brutetree from NearestNeighbors might do the job as it is basically brute force NN with a very simple vector of vector structure. For low dimensionalities may be enough...
Hi, I just wanted to try out this package. I've used NearestNeigbors.jl before, but need to be able to add and remove points from the tree efficiently.
I want to start with an empty index (nvectors=0) and build it up, but got stuck on this assert:
IVFADC.jl/src/index.jl
Line 119 in b2a9e0b
Is there a way to do this? I don't know much about these indices and if this would be the right package for the job.
The text was updated successfully, but these errors were encountered: