[
https://issues.apache.org/jira/browse/LUCENE-8647?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16745452#comment-16745452
]
Sambhav Kothari commented on LUCENE-8647:
-----------------------------------------
Example implementation of a subclass -
{code:java}
/**
* A special explanation class that stores feature scores for a specific doc
that a model is scoring.
*/
public class ModelExplanation extends Explanation {
private Map<String, Float> featureImportance = new HashMap<>();
private Map<String, Float> featureValues = new HashMap<>();
public Map<String, Float> getFeatureImportance() {
return featureImportance;
}
public void setFeatureImportance(Map<String, Float> featureImportance) {
this.featureImportance = featureImportance;
}
public Map<String, Float> getFeatureValues() {
return featureValues;
}
public void setFeatureValues(Map<String, Float> featureValues) {
this.featureValues = featureValues;
}
@Override
public LinkedHashMap<String, Object> toMap() {
LinkedHashMap<String, Object> data = super.toMap();
data.put("featureImportance", getFeatureImportance());
data.put("featureValues", getFeatureValues());
return data;
}
}
{code}
> Create a generic method in the Explanation class to get explanation attributes
> ------------------------------------------------------------------------------
>
> Key: LUCENE-8647
> URL: https://issues.apache.org/jira/browse/LUCENE-8647
> Project: Lucene - Core
> Issue Type: Improvement
> Components: core/search
> Affects Versions: 7.5
> Reporter: Sambhav Kothari
> Priority: Minor
>
> Initial proposal and discussion can be found here -
>
> [http://mail-archives.apache.org/mod_mbox/lucene-java-user/201901.mbox/%3CCADitDkmnv4Vup5VG2ZwLUOv5VM=itu-y8eugxxmtoppzpcy...@mail.gmail.com%3E]
>
> Currently the Explanation class has a toString method, but it doesn't seem to
> have any method to output the explanation as a NamedList. Part of the reason
> is that NamedList is a Solr only concept and it cannot exist in lucene. But
> this leads to utility functions like
> ([https://github.com/apache/lucene-solr/blob/1d85cd783863f75cea133fb9c452302214165a4d/solr/core/src/java/org/apache/solr/response/transform/ExplainAugmenterFactory.java#L115]
> and
> [https://github.com/apache/lucene-solr/blob/1d85cd783863f75cea133fb9c452302214165a4d/solr/core/src/java/org/apache/solr/util/SolrPluginUtils.java#L400-L466]
> which assume a particular structure about explanations.
>
> Explanation should instead have a toMap method which returns a Map of
> key-value pairs which can then be converted to a NamedList in Solr. This can
> be exceptionally useful if we have a class inheriting from Explanation that
> adds more attributes.
>
> For example for Learning to Rank model explanations we can have a
> ModelExplanation class which adds doc-specific feature scores and other
> machine learning model explanations to the output. It also allows the
> ExplainAugmenter and other classes to have a more generic way of getting an
> Explanation's data in a structured format.
>
> The way it currently does so is very ugly -
> [https://github.com/apache/lucene-solr/blob/1d85cd783863f75cea133fb9c452302214165a4d/solr/core/src/java/org/apache/solr/response/transform/ExplainAugmenterFactory.java#L115]
>
> This also implies that we shouldn't keep Explanation as a final class
> anymore. Storing and parsing things like feature values in the explanation
> description doesn't seem like the right way.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
