Ranker scores shouldn't be discarded

[MichelBartels]

Is your feature request related to a problem? Please describe.
The reason for using a ranker is that the scores should be more accurate than the retriever scores. For this reason, in that use case you would also want to use the ranker scores later on instead of the retriever score (e.g. when defining a threshold). You currently cannot do that as the ranker scores aren't saved anywhere (see here) as the reader only updates the order and not the score. Having the order of documents not dependent on the scores is also unintuitive.

Describe the solution you'd like
Replace retriever scores with ranker scores.

Describe alternatives you've considered
Having an additional field to save ranker scores.

[MichelBartels]

It seems I have overlooked #2048 which should solve this issue.

[julian-risch]

Hi @MichelBartels the corresponding PR #2048 is about to be merged, feel free to test it out and let me know if there are any problems. 🙂

[MichelBartels]

Hi @julian-risch, I have already used the PR and can confirm that it is solving this issue.