prateekpatelsc

@yurymalkov : any comments on the quality of the results if the index could be possibly stored in hnswlib as well

@yurymalkov : sorry may be my question was not clear. I meant to ask that hnsw-ivf uses quantizers (from a quick look at code), i was curious if that causes...

Thanks for clarification . So this implies that searching for a larger topK (top 1000) and then taking the smallest top100 among them is a different result than searching for...

What i was trying was : build index with ef_constuction and fixed M , not if topM with ef=ef_consruction recall

I am not talking about modifying any vector . I want to build an index that also supports filtering i.e give me topK neighbors which have tags in [.. tag...

may be i am not phrasing it properly. If you get knn and then pop , this is just post filtering . Are you suggesting this ? because this would...

@yurymalkov : one issue is inverted index could result in large (millions of ids, index could be hundred million) to create large allowlist . I am wondering if its reasonable...

Trying to understand the difference between the 2 approach : 1) treat items not in allowlist as deleted 2) Compute a large distance for items not in allowlist In the...

Is this of because stuck in some local minima where no neighbors are improving distance and the search queue is empty ?

Also is this somehow related to data dimensionality as well ? for example for a fixed M and search ef , ef_construction param , the chances of running into these...