I think that this optimization was introduced because `advanceShallow` may
advance skip lists and then never decode a block of postings. But actually
`IndexInput#seek` is cheap, including on `NIOFSDirectory`. So let's seek
immediately?
* #7820: add initial (failing) test case exposing the bug in CheckIndex
* #7820: initial fix to CheckIndex to detect 'on load' corruption of older segments files
* exclamation point police
* tidy
* add missing @Override in new test case
* fold feedback: merge SIS-loading logic into single for-loop; rename sis -> lastCommit
* tidy
* sidestep segments.gen file as well in case we are reading very old index
* tidy
* always load last commit point first so if it has an issue that will not be masked by issues with older commit points
When requesting for k >= numVectors, it doesn't make sense to go through the HNSW graph. Even without a user supplied filter, we should not explore the HNSW graph if it contains fewer than k vectors.
One scenario where we may still explore the graph if k >= numVectors is when not every document has a vector and there are deleted docs. But, this commit significantly improves things regardless.
This was found by `testRandomExceptions()`: if an exception occurs when opening
the meta file, then the `rawVectorsReader` that is passed to the constructor
never gets closed.
Prevent contention on the ReentrantReadWriteLock at search time by
ensuring that `commit` and `purgeCache` are never both trying to
hold the write lock.
This commit refactors the usage of the deprecated System::runFinalization in tests and benchmarks, so that narrowly targeted suppressions can be added.
Currently the HNSW codec does too many things, it not only indexes vectors, but stores them and determines how to store them given the vector type.
This PR extracts out the vector storage into a new format `Lucene99FlatVectorsFormat` and adds new base class called `FlatVectorsFormat`. This allows for some additional helper functions that allow an indexing codec (like HNSW) take advantage of the flat formats.
Additionally, this PR refactors the new `Lucene99ScalarQuantizedVectorsFormat` to be a `FlatVectorsFormat`.
Now, `Lucene99HnswVectorsFormat` is constructed with a `Lucene99FlatVectorsFormat` and a new `Lucene99HnswScalarQuantizedVectorsFormat` that uses `Lucene99ScalarQuantizedVectorsFormat`
* Skip document by docValues
*When the queue is full with only one Comparator, we could better tune the maxValueAsBytes/minValueAsBytes. For instance, if the sort is ascending and bottom value is 5, we will use a range on [MIN_VALUE, 4].
---------
Co-authored-by: Adrien Grand <jpountz@gmail.com>
### Description
When using cosine similarity, the `ScalarQuantizer` normalizes vectors when calculating quantiles and `ScalarQuantizedRandomVectorScorer` normalizes query vectors before scoring them, but `Lucene99ScalarQuantizedVectorsWriter` does not normalize the vectors prior to quantizing them when producing the quantized vectors to write to disk. This PR normalizes vectors prior to quantizing them when writing them to disk.
Recall results on my M1 with the `glove-100-angular` data set (all using `maxConn`: 16, `beamWidth` 100, `numCandidates`: 100, `k`: 10, single segment):
| Configuration | Recall | Average Query Duration |
|---------------|-------|-----------------|
| Pre-patch no quantization | 0.78762 | 0.68 ms |
| Pre-patch with quantization | 8.999999999999717E-5 | 0.45 ms |
| Post-patch no quantization | 0.78762 | 0.70 ms |
| Post-patch with quantization | 0.66742 | 0.66 ms |
Clean-up from adding the Lucene99PostingsFormat in https://github.com/apache/lucene/pull/12741
These test cases were moved to Lucene99 dir and I forgot to copy the unmodified versions for the backward_codecs.lucene90
Both testEuclidean and testExplain have vectors that result
in equal scores. Since we no longer tie break on vector ordinal
as it doesn't make sense when building the graph, the vectors returned
might be slightly different. This commit fixes the flaky nature of the
test.
I noticed while testing lower dimensionality and quantization, we would explore the HNSW graph way too much. I was stuck figuring out why until I noticed the searcher checks for distance equality (not just if the distance is better) when exploring neighbors-of-neighbors. This seems like a bad heuristic, but to double check I looked at what nmslib does. This pointed me back to this commit: nmslib/nmslib#106
Seems like this performance hitch was discovered awhile ago :).
This commit adjusts HNSW to only explore the graph layer if the distance is actually better.
Adrien Grand
Shibi Balamurugan
Michael McCandless
Jeff Zemerick
Benjamin Trent
Uwe Schindler
Alan Woodward
Davis Cook
Juan Manuel Caicedo Carvajal
zhouhui
Greg Miller
Dzung Bui
Jakub Slowinski
Shubham Chaudhary
Stefan Vodita
Chris Hegarty
Benjamin Trent
Patrick Zhai
Robert Muir
gf2121
Egor Potemkin
Zhang Chao
Lu Xugang
Kevin Rosendahl
Dawid Weiss