Julian Risch 84c34295a1
Re-ranking component for document search without QA (#1025)
* Adding ranker similar to retriever and reader

* Sort documents according to query-document similarity scores

* Reranking and model training runs for small example

* Added EvalRanker node

* Calculate recall@k in EvalRetriever and EvalRanker nodes

* Renaming EvalRetriever to EvalDocuments and EvalReader to EvalAnswers

* Added mean reciprocal rank as metric for EvalDocuments

* Fix bug that appeared when ranking documents with same score

* Remove commented code for unimplmented eval() of Ranker node

* Add documentation of k parameter in EvalDocuments

* Add Ranker docu and renaming top_k param
2021-05-31 15:31:36 +02:00
..
2021-03-18 18:00:21 +01:00
2021-02-15 11:15:54 +01:00
2021-04-22 16:45:29 +02:00
2021-04-08 14:05:33 +02:00
2021-04-29 18:19:20 +02:00
2021-04-29 18:19:20 +02:00