Hmm yes it looked that way to me, thanks Adrien. Maybe there is an
opportunity in the non scoring case to track these and save some effort.
On Wed, Nov 14, 2018, 9:10 AM Adrien Grand Hi Mike,
>
> The best approach would be to run the query again indeed since the
> join is performed in a lossy way
Hi Mike,
The best approach would be to run the query again indeed since the
join is performed in a lossy way: we record score contributions but
not which documents contributed to the score. For instance this is
what the inner_hits feature of Elasticsearch does when used with
query-time joins.
On F