JunRuiLee commented on issue #97:
URL:
https://github.com/apache/paimon-vector-index/issues/97#issuecomment-5652259862
Both clarifications are in the issue text now, along with the reference note.
1. The roadmap no longer offers top-K plus post-filtering as a fallback. It
says
an unsupported family fails loud with a distinguishable error; that top-K
plus
post-filtering is not equivalent, since with 100 vectors in the band and
K = 10 at most 10 can come back and that holds for exact search too; that
a
caller needing the result set to equal the predicate needs an execution
path
preserving it, namely an exhaustive scan; and that a capped fallback is
something a caller opts into explicitly. #99's description carries the
same
wrong wording and is being corrected with it.
2. Added, right after the coverage table: exactness and completeness are
separate
properties. IVF-Flat computes exact distances for the rows it visits, but
a
matching row in an unprobed list is still missed whenever `nprobe <
nlist`.
Removing K removes truncation; it does not make any IVF family exhaustive.
3. On the reference: correct. This crate's IVF-RQ is RaBitQ
(`core/src/rq.rs`),
and `docs/ivf-rq.html` already cross-checks it against Faiss's
`IndexIVFRaBitQFastScan`. The table cell named the additive-quantizer
family,
which invites exactly the confusion you describe; it now points at the
`IndexIVF` range contract generally, with a note that what carries over
from
Faiss is that contract and not the quantizer.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
