-
Notifications
You must be signed in to change notification settings - Fork 1.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
feat: add support for BM25Retriever
in InMemoryDocumentStore
#3561
Conversation
Proposed Changes:As discussed in #3447, I made the To make the How did you test it?As you can see from this notebook, this implementation works fine for Haystack Tutorial 1. For more proper tests, I need help:
@ZanSara feel free to jump in! 🙂 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Fantastic! Thank you for the great work! 🚀
Regarding tests, I believe they should come from two angles:
-
First we need unit tests as part of the document store tests. I'm sure there are no specific InMemoryDocumentStore tests now, but please add a few in the main docstore tests
test_document_store.py
, and parametrize them to run onmemory
only. @masci will take care of them shortly after we merge this PR -
Then we can test them from BM25Retriever side. I think it's sufficient to add
(elasticsearch, memory)
to this fixture: https://github.com/deepset-ai/haystack/blob/main/test/nodes/test_retriever.py#L33-L53- Incidentally this would be a good time to fix the parametrization of these tests and replace
elasticsearch
withbm25
to representBM25Retriever
in tests. The switch needs to be done:- Here:
Line 723 in 4dfddf0
@pytest.fixture(params=["es_filter_only", "elasticsearch", "dpr", "embedding", "tfidf", "table_text_retriever"]) - Here: https://github.com/deepset-ai/haystack/blob/main/test/conftest.py#L789
- And in every test that uses the
retriever
andretriever_with_docs
fixture 🥲 Please don't feel obliged to do this, I'll take care of it if it's too much work 👍
- Here:
- Incidentally this would be a good time to fix the parametrization of these tests and replace
I added some document store tests. As usual, there is room for improvement... @ZanSara 😃
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Very nice! Thank you for taking care of the fixture renaming. There's one last thing that we should probably fix and then it can be merged 😊
@ZanSara the CI is failing in a strange way: Any ideas? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sorry to hold this one again but indeed you found a super valid corner case by setting use_bm25=True
as default in tests. Good to see the test suite being really useful for once 😄
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great! Thank you so much! 🚀
@deepset-ai/documentation Let's update the docs to reflect this change |
* Fix docstrings for DocumentStores * Fix docstrings for AnswerGenerator * Fix docstrings for Connector * Fix docstrings for DocumentClassifier * Fix docstrings for LabelGenerator * Fix docstrings for QueryClassifier * Fix docstrings for Ranker * Fix docstrings for Retriever and Summarizer * Fix docstrings for Translator * Fix docstrings for Pipelines * Fix docstrings for Primitives * Fix Python code block spacing * Add line break before code block * Fix code blocks * fix: discard metadata fields if not set in Weaviate (#3578) * fix weaviate bug in returning embeddings and setting empty meta fields * review comment * Update unstable version and openapi schema (#3584) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> * fix: Flatten `DocumentClassifier` output in `SQLDocumentStore`; remove `_sql_session_rollback` hack in tests (#3273) * first draft * fix * fix * move test to test_sql * test: add test to check id_hash_keys is not ignored (#3577) * refactor: Generate JSON schema when missing (#3533) * removed unused script * print info logs when generating openapi schema * create json schema only when needed * fix tests * Remove leftover Co-authored-by: ZanSara <sarazanzo94@gmail.com> * move milvus tests to their own module (#3596) * feat: store metadata using JSON in SQLDocumentStore (#3547) * add warnings * make the field cachable * review comment * Pin faiss-cpu as 1.7.3 seems to have problems (#3603) * Update Haystack imports (#3599) * Update Python version (#3602) * fix: `ParsrConverter` fails on pages without text (#3605) * try to fix bug * remove print * leftover * refactor: update Squad data (#3513) * refractor the to_squad data class * fix the validation label * refractor the to_squad data class * fix the validation label * add the test for the to_label object function * fix the tests for to_label_objects * move all the test related to squad data to one file * remove unused imports * revert tiny_augmented.json Co-authored-by: ZanSara <sarazanzo94@gmail.com> * Url fixes (#3592) * add 2 example scripts * fixing faq script * fixing some urls * removing example scripts * black reformatting * add labeler to the repo (#3609) * convert eval metrics to python float (#3612) * feat: add support for `BM25Retriever` in `InMemoryDocumentStore` (#3561) * very first draft * implement query and query_batch * add more bm25 parameters * add rank_bm25 dependency * fix mypy * remove tokenizer callable parameter * remove unused import * only json serializable attributes * try to fix: pylint too-many-public-methods / R0904 * bm25 attribute always present * convert errors into warnings to make the tutorial 1 work * add docstrings; tests * try to make tests run * better docstrings; revert not running tests * some suggestions from review * rename elasticsearch retriever as bm25 in tests; try to test memory_bm25 * exclude tests with filters * change elasticsearch to bm25 retriever in test_summarizer * add tests * try to improve tests * better type hint * adapt test_table_text_retriever_embedding * handle non-textual docs * query only textual documents * Incorporate Reviewer feedback * refactor: replace `torch.no_grad` with `torch.inference_mode` (where possible) (#3601) * try to replace torch.no_grad * revert erroneous change * revert other module breaking * revert training/base * Fix docstrings for DocumentStores * Fix docstrings for AnswerGenerator * Fix docstrings for Connector * Fix docstrings for DocumentClassifier * Fix docstrings for LabelGenerator * Fix docstrings for QueryClassifier * Fix docstrings for Ranker * Fix docstrings for Retriever and Summarizer * Fix docstrings for Translator * Fix docstrings for Pipelines * Fix docstrings for Primitives * Fix Python code block spacing * Add line break before code block * Fix code blocks * Incorporate Reviewer feedback Co-authored-by: Massimiliano Pippi <mpippi@gmail.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com> Co-authored-by: Stefano Fiorucci <44616784+anakin87@users.noreply.github.com> Co-authored-by: Julian Risch <julian.risch@deepset.ai> Co-authored-by: ZanSara <sarazanzo94@gmail.com> Co-authored-by: Espoir Murhabazi <espoir.mur@gmail.com> Co-authored-by: Tuana Celik <tuana.celik@deepset.ai> Co-authored-by: tstadel <60758086+tstadel@users.noreply.github.com>
Related Issues
Only a first draft...
Checklist