Hi Ashis,
Short answer: No, i don't think it's possible.
I'm considering as well extending solr to allow plugging in features from
outside, but it will require time because at the moment the features can
see only the current document processed, while to do that ideally you want
to process in one shot all the documents that you are reranking...
In your specific case, could you index the docembedding with the document,
send the query embedding as external feature information (efi) and
implement a Java feature that combines them?
On Wed, Apr 24, 2019, 09:12 Ashis Tripathy wrote:
> Hi ,
>
> Hope you are doing well!!!
>
> I am currently trying to implement Learn To Rank in SOLR 7.5. I have gone
> through the blogs and documentation page to follow the steps. However the
> current process looks like i need to write SOLR query in a json file to
> upload the same to Feature Store.
>
> But the problem is , some of the features which I want to create can't be
> created using SOLR query syntax. For example: Document embeddings and query
> embeddings using Doc2Vec model in python.
>
> Is there any way i can generate the features outside SOLR by passing the
> query and the top-K result set to a method written in python and then
> rerank the results using the trained model.
>
> Any help will be highly appreciated as this thing is blocking me completely
> from going ahead.
>
> few useful links which i found:
> https://github.com/airalcorn2/Solr-LTR
> https://lucene.apache.org/solr/guide/7_5/learning-to-rank.html
>
> please let me know in case you need any other details about the issue.
>
> Regards
> Ashis
>
