Add a custom vector_script engine. #6
Closed
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
jtibshirani
force-pushed
the
vector-script-query
branch
from
2817ae6
to
731667c
Compare
This calculation omits some safety checks around vector length for simplicity (these weren't found to have a significant effect on performance).
jtibshirani
force-pushed
the
vector-script-query
branch
from
731667c
to
ef8c261
Compare
jtibshirani
pushed a commit
that referenced
this pull request
Mar 11, 2021
…astic#69765) Previously we did not resolve the attributes recursively which meant that if a field or expression was re-aliased multiple times (through multiple levels of subqueries), the aliases were only resolved one level down. This led to failed query translation because `ReferenceAttribute`s were pointing to non-existing attributes during query translation. For example the query ```sql SELECT i AS j FROM ( SELECT int AS i FROM test) ORDER BY j ``` failed during translation because the `OrderBy` resolved the `j` ReferenceAttribute to another `i` ReferenceAttribute that was later removed by an Optimization: ``` OrderBy[[Order[j{r}#4,ASC,LAST]]] ! OrderBy[[Order[i{r}#2,ASC,LAST]]] \_Project[[j]] = \_Project[[j]] \_Project[[i]] ! \_EsRelation[test][date{f}#6, some{f}#7, some.string{f}#8, some.string..] \_EsRelation[test][date{f}#6, some{f}#7, some.string{f}#8, some.string..] ! ``` By resolving the `Attributes` recursively both `j{r}` and `i{r}` will resolve to `test.int{f}` above: ``` OrderBy[[Order[test.int{f}elastic#22,ASC,LAST]]] = OrderBy[[Order[test.int{f}elastic#22,ASC,LAST]]] \_Project[[j]] = \_Project[[j]] \_Project[[i]] ! \_EsRelation[test][date{f}#6, some{f}#7, some.string{f}#8, some.string..] \_EsRelation[test][date{f}#6, some{f}#7, some.string{f}#8, some.string..] ! ``` The scope of recursive resolution depends on how the `AttributeMap` is constructed and populated. Fixes elastic#67237
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.
This helps quantify the performance difference due to scripting. Results on
random-s-100-angular:baselineis the current approach to vector scoring (in elasticsearch 7.5)vector_queryruns the dedicated query which avoids scripting altogether (Add a dedicated vector_score query. #4)vector_scriptuses the custom script engine from this PR (commit 2212966)vector_script_directuses the custom script engine from this PR, but computes the similarity directly instead of usingCosineSimilarity(commit 65d73f1)direct_docvaluesbuilds onvector_script_directand accesses docvalues directly as opposed to usingLeafDocLookup(commit ef8c261)