[Bug] query_namespaces can handle single result #421
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
jhamon
commented
Dec 6, 2024
| @@ -145,22 +147,7 @@ def test_query_namespaces(self, idx): | |||
| assert len(results6.matches) == 0 | |||
| assert results6.usage.read_units > 0 | |||
|
|
|||
| def test_invalid_top_k(self, idx): | |||
There was a problem hiding this comment.
Don't need this test anymore since top_k of 1 is now valid.
jhamon
requested review from
austin-denoble,
haruska,
rohanshah18 and
aulorbe
austin-denoble
approved these changes
Dec 6, 2024
rohanshah18
approved these changes
Dec 6, 2024
jhamon
added a commit
that referenced
this pull request
Dec 13, 2024
In order to merge results across multiple queries, the SDK must know which similarity metric an index is using. For dotproduct and cosine indexes, a larger score is better while for euclidean a smaller score is better. Unfortunately the data plane API does not currently expose the metric type and a separate call to the control plane to find out seems undesirable from a resiliency and performance perspective. As a workaround, in the initial implementation of `query_namespaces` the SDK would infer the similarity metric needed to merge results by seeing whether the scores of query results were ascending or descending. This worked well, but imposes an implicit limitation that there must be at least 2 results returned. We initially believed this would not be a problem but have since learned that applications using filtering can sometimes filter out all or most results. So an approach that has the user explicitly telling the SDK what similarity metric is being used is preferred to handle these edge cases with 1 or 0 results. - Add a required kwarg to `query_namespaces` to specify the index similarity metric. - Modify `QueryResultsAggregator` to use this similarity metric, and strip out code that was involved in inferring whether results were ascending or descending. - Adjust integration tests to pass new metric kwarg. Except for adding the new kwarg, query_namespaces integration tests did not need to change which indicates the underlying behavior is still working as before. - [x] Bug fix (non-breaking change which fixes an issue)
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.
Problem
In order to merge results across multiple queries, the SDK must know which similarity metric an index is using. For dotproduct and cosine indexes, a larger score is better while for euclidean a smaller score is better. Unfortunately the data plane API does not currently expose the metric type and a separate call to the control plane to find out seems undesirable from a resiliency and performance perspective.
As a workaround, in the initial implementation of
query_namespacesthe SDK would infer the similarity metric needed to merge results by seeing whether the scores of query results were ascending or descending. This worked well, but imposes an implicit limitation that there must be at least 2 results returned.We initially believed this would not be a problem but have since learned that applications using filtering can sometimes filter out all or most results. So an approach that has the user explicitly telling the SDK what similarity metric is being used is preferred to handle these edge cases with 1 or 0 results.
Solution
query_namespacesto specify the index similarity metric.QueryResultsAggregatorto use this similarity metric, and strip out code that was involved in inferring whether results were ascending or descending.Type of Change