Replies: 4 comments
-
>>> import faiss
>>> index = faiss.index_factory(128, "HNSW32,Flat")
>>> index.hnsw.nb_neighbors(1)
32
>>> index.hnsw.nb_neighbors(0)
64 |
Beta Was this translation helpful? Give feedback.
-
@KinglittleQ Thank you for the reply. |
Beta Was this translation helpful? Give feedback.
-
This does not seems to be correct. With M=32, one vector will use |
Beta Was this translation helpful? Give feedback.
-
Ok, I read the paper: "Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs". In chapter 4.2.3 it listed memory cost for index is
If we ignore the term Though I am not sure how close the implementation is according to the paper. |
Beta Was this translation helpful? Give feedback.
-
Summary
What is the meaning of
x
inHNSWx,Flat
? I'm confused of its space footprint4*d + x * M * 2 * 4
. I know M is the maximum number of degree on each layer and d is the dimensionality of vectors to index. But I could not get the meaning ofx
. Is it a number of layers that the algorithm is producing?Beta Was this translation helpful? Give feedback.
All reactions