[
https://issues.apache.org/jira/browse/SOLR-16788?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Anna updated SOLR-16788:
------------------------
Description:
Efi features are not correctly printed when these conditions persist:
* Asking for logging:
{code:java}
fl=[features format=dense efi.myEfiFeature=2]{code}
* Default feature store used (no feature store passed)
* Asking for reranking:
{code:java}
{!ltr model=my_model efi.myEfiFeature=3}{code}
* Different efis values are passed between logging and reranking:
efi.myEfiFeature=2 vs efi.myEfiFeature=3
The efis logged are computed based on the reranking and not the logging.
Example of response:
{code:java}
"docs": [
{
"id": "41",
"score": 4999.0,
"[features]": "test_efi_0=4996.0,test_efi_1=0.0,efi_feature=3.0"
},
{
"id": "80",
"score": 4955.0,
"[features]": "test_efi_0=4952.0,test_efi_1=0.0,efi_feature=3.0"
},
...{code}
I would expect to have efi_feature=2.0 in this scenario.
This is due to the
*org.apache.solr.ltr.response.transform.LTRFeatureLoggerTransformerFactory.FeatureTransformer#setupRerankingQueriesForLogging*
method, which is not assigning the _transformerExternalFeatureInfo_ parameter
to the
_LTRScoringQuery_ if we have done reranking and the _transformerFeatureStore_
is null (not passed in fl at query time).
was:
Efi features are not correctly printed when these conditions persist:
# Asking for logging:
{code:java}
fl=[features format=dense efi.myEfiFeature=2]{code}
# Default feature store used (no feature store passed)
# Asking for reranking:
{code:java}
{!ltr model=my_model efi.myEfiFeature=3}{code}
# Different efis values are passed between logging and reranking:
efi.myEfiFeature=2 vs efi.myEfiFeature=3
The efis logged are computed based on the reranking and not the logging.
Example of response:
{code:java}
"docs": [
{
"id": "41",
"score": 4999.0,
"[features]": "test_efi_0=4996.0,test_efi_1=0.0,efi_feature=3.0"
},
{
"id": "80",
"score": 4955.0,
"[features]": "test_efi_0=4952.0,test_efi_1=0.0,efi_feature=3.0"
},
...{code}
I would expect to have efi_feature=2.0 in this scenario.
This is due to the
*org.apache.solr.ltr.response.transform.LTRFeatureLoggerTransformerFactory.FeatureTransformer#setupRerankingQueriesForLogging*
method, which is not assigning the _transformerExternalFeatureInfo_ parameter
to the
_LTRScoringQuery_ if we have done reranking and the _transformerFeatureStore_
is null (not passed in fl at query time).
> Bug in logging efi features
> ---------------------------
>
> Key: SOLR-16788
> URL: https://issues.apache.org/jira/browse/SOLR-16788
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Components: contrib - LTR
> Affects Versions: 9.2.1
> Reporter: Anna
> Priority: Minor
>
> Efi features are not correctly printed when these conditions persist:
> * Asking for logging:
> {code:java}
> fl=[features format=dense efi.myEfiFeature=2]{code}
> * Default feature store used (no feature store passed)
> * Asking for reranking:
> {code:java}
> {!ltr model=my_model efi.myEfiFeature=3}{code}
> * Different efis values are passed between logging and reranking:
> efi.myEfiFeature=2 vs efi.myEfiFeature=3
> The efis logged are computed based on the reranking and not the logging.
> Example of response:
> {code:java}
> "docs": [
> {
> "id": "41",
> "score": 4999.0,
> "[features]": "test_efi_0=4996.0,test_efi_1=0.0,efi_feature=3.0"
> },
> {
> "id": "80",
> "score": 4955.0,
> "[features]": "test_efi_0=4952.0,test_efi_1=0.0,efi_feature=3.0"
> },
> ...{code}
> I would expect to have efi_feature=2.0 in this scenario.
>
> This is due to the
> *org.apache.solr.ltr.response.transform.LTRFeatureLoggerTransformerFactory.FeatureTransformer#setupRerankingQueriesForLogging*
> method, which is not assigning the _transformerExternalFeatureInfo_
> parameter to the
> _LTRScoringQuery_ if we have done reranking and the _transformerFeatureStore_
> is null (not passed in fl at query time).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
