feat: Weights and score normalization for DocumentJoiner with reciprocal rank fusion

[robpasternak]

Related Issues

• fixes Weights and score normalization for DocumentJoiner with reciprocal rank fusion - 2.x #6617 by adding weighting and score normalization for the DocumentJoiner node with reciprocal rank fusion.

Proposed Changes:

• Add weighting for DocumentJoiner node with reciprocal rank fusion.
• Add score normalization for DocumentJoiner node with reciprocal rank fusion.

How did you test it?

• Testing pending.

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.
• I documented my code
• I ran pre-commit hooks and fixed any issue

[coveralls]

Pull Request Test Coverage Report for Build 7641580805

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.006%) to 88.88%

---

Totals
Change from base Build 7641576479:	0.006%
Covered Lines:	4524
Relevant Lines:	5090

---

💛 - Coveralls

[robpasternak]

I just did some simple tests by hand, and it seems to work! Test code and output below. Will now mark it ready for review.

(Note: I didn't see a reasonable spot to include this as a test in the actual repo itself--all the DocumentJoiner tests seem to deal with much more fundamental stuff--which is why I did the tests by hand. If you'd like me to somehow incorporate tests in the repo let me know.)

Test code

from haystack.components.joiners import DocumentJoiner
from haystack import Document

def make_docs(
    num_docs: int = 6,
):
    out_docs = []
    for i in range(num_docs):
        out_docs.append(Document(content=f"doc{i}"))
    return out_docs

if __name__ == '__main__':
    docs = make_docs(6)
    docs_2 = [docs[0], docs[4], docs[2], docs[5], docs[1]]
    document_lists = [docs, docs_2]

    joiner_1 = DocumentJoiner(
        join_mode = "reciprocal_rank_fusion",
        weights = [0.5,0.5],
    )

    joiner_2 = DocumentJoiner(
        join_mode = "reciprocal_rank_fusion",
        weights = [7,7],
    )

    joiner_3 = DocumentJoiner(
        join_mode = "reciprocal_rank_fusion",
        weights = [0.7,0.3],
    )

    joiner_4 = DocumentJoiner(
        join_mode = "reciprocal_rank_fusion",
        weights = [0.6,0.4],
    )

    joiner_5 = DocumentJoiner(
        join_mode = "reciprocal_rank_fusion",
        weights = [1,0],
    )

    joiners = [joiner_1, joiner_2, joiner_3, joiner_4, joiner_5]
    
    for index, joiner in enumerate(joiners):
        print(f'JOINER {index+1}:')
        join_results = joiner.run(documents = document_lists)
        for doc in join_results['documents']:
            print(doc.content, doc.score)
        print('')

Output

JOINER 1:
doc0 1.0
doc2 0.9682539682539681
doc1 0.961166253101737
doc4 0.961166253101737
doc5 0.938683712121212
doc3 0.4765625

JOINER 2:
doc0 1.0
doc2 0.9682539682539681
doc1 0.961166253101737
doc4 0.961166253101737
doc5 0.938683712121212
doc3 0.4765625

JOINER 3:
doc0 0.9999999999999998
doc1 0.9702481389578161
doc2 0.9682539682539681
doc4 0.9520843672456575
doc5 0.9329071969696969
doc3 0.6671874999999999

JOINER 4:
doc0 0.9999999999999998
doc2 0.9682539682539681
doc1 0.9657071960297765
doc4 0.9566253101736971
doc5 0.9357954545454545
doc3 0.5718749999999999

JOINER 5:
doc0 1.0
doc1 0.9838709677419354
doc2 0.9682539682539681
doc3 0.953125
doc4 0.9384615384615385
doc5 0.9242424242424242

[robpasternak]

Oh and another comment: we've previously talked about making the score normalization optional, rather than obligatory. This PR keeps the score normalization as obligatory, like in the v1.x JoinDocuments node. I did this because I thought the decision about how to implement any optionality would be best left to Core Engineering folks. But I'm happy to offer hot takes and/or implement changes if you'd like 🙂

[vblagoje]

Hey @robpasternak @anakin87 seems ok. The. only slight concern I have is document_lists being zero len blowing up pipeline run with ZeroDivisionError. No other code path in this component addresses this, so we can leave it for more consistent updates across all components, right @anakin87

So in conclusion, a slightly longer release note blurb explaining this to users would be appreciated. I'll share our guidelines internally.

[vblagoje]

Let's write a bit longer release note please

[vblagoje]

Also @robpasternak would you also please just add the test you wrote to test_document_joiner.py?

[vblagoje]

🚢

[robpasternak]

Oh thanks for jumping on the changes @vblagoje! Apologies for not doing them sooner myself, was gonna do them today but failed to communicate that. Anyway, thanks again!

[vblagoje]

No worries @robpasternak it took me literally a few minutes to update, all credit is yours 🚀