Add CompressionLevel Calculation for PQ #2200
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jmazanec15
added
Enhancements
jmazanec15
requested review from
heemin32,
navneet1v,
VijayanB,
vamshin,
naveentatikonda,
junqiu-lei,
martin-gaievski,
ryanbogan and
luyuncheng
as code owners
jmazanec15
force-pushed
the
pq-compression-level-fix
branch
4 times, most recently
from
f2d7c89 to
907f1ec
Compare
should we allow more compression level? |
navneet1v
reviewed
Oct 10, 2024
| @@ -29,6 +29,8 @@ public enum CompressionLevel { | |||
| x16(16, "16x", new RescoreContext(3.0f, false), Set.of(Mode.ON_DISK)), | |||
| x32(32, "32x", new RescoreContext(3.0f, false), Set.of(Mode.ON_DISK)); | |||
|
|
|||
| public static final CompressionLevel MAX_COMPRESSION_LEVEL = CompressionLevel.x32; | |||
There was a problem hiding this comment.
should we have 64 as a max compression level? I don't have a solid point to have 64 but I think have 1 more extra compression is always good.
There was a problem hiding this comment.
I think that makes sense. I guess for 64x, default for all dimensions should probably be 5x.
navneet1v
reviewed
Oct 10, 2024
heemin32
reviewed
Oct 10, 2024
heemin32
previously approved these changes
Oct 10, 2024
Currently, for product quantization, we set the calculated compression level to NOT_CONFIGURED. The main issue with this is that if a user sets up a disk-based index with PQ, no re-scoring will happen by default. This change adds the calculation so that the proper re-scoring will happen. The formula is fairly straightforward => actual compression = (d * 32) / (m * code_size). Then, we round to the neareste compression level (because we only support discrete compression levels). One small issue with this is that if PQ is configured to have compression > 32x, the value will be 32x. Functionally, the only issue will be that we may not be as aggressive on oversampling for on disk mode. Signed-off-by: John Mazanec <[email protected]>
Signed-off-by: John Mazanec <[email protected]>
jmazanec15
force-pushed
the
pq-compression-level-fix
branch
from
907f1ec to
78915d1
Compare
navneet1v
approved these changes
Oct 16, 2024
heemin32
approved these changes
Oct 16, 2024
opensearch-trigger-bot bot
pushed a commit
that referenced
this pull request
Oct 16, 2024
Currently, for product quantization, we set the calculated compression level to NOT_CONFIGURED. The main issue with this is that if a user sets up a disk-based index with PQ, no re-scoring will happen by default. This change adds the calculation so that the proper re-scoring will happen. The formula is fairly straightforward => actual compression = (d * 32) / (m * code_size). Then, we round to the neareste compression level (because we only support discrete compression levels). One small issue with this is that if PQ is configured to have compression > 64x, the value will be 64x. Functionally, the only issue will be that we may not be as aggressive on oversampling for on disk mode. Signed-off-by: John Mazanec <[email protected]> (cherry picked from commit 228aead)
|
@jmazanec15 did we do any benchmarks which suggest how much improvement we will get with PQ based rescoring? |
|
@navneet1v we did here: #1779 (comment) |
Thanks. I completely forgot about that. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Currently, for product quantization, we set the calculated compression level to NOT_CONFIGURED. The main issue with this is that if a user sets up a disk-based index with PQ, no re-scoring will happen by default.
This change adds the calculation so that the proper re-scoring will happen. The formula is fairly straightforward =>
actual compression = (d * 32) / (m * code_size). Then, we round to the neareste compression level (because we only support discrete compression levels).
One small issue with this is that if PQ is configured to have compression > 32x, the value will be 32x. Functionally, the only issue will be that we may not be as aggressive on oversampling for on disk mode.
Check List
--signoff.By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.