raphaelsty / cherche Goto Github PK

hi, please is it possible to filter the search

https://github.com/raphaelsty/cherche/blob/6aabe70853d2486ee1972e1689290fdce8b72587/cherche/index/faiss_index.py#L107C13-L107C64

I'm using a cherche pipline built of a tfidf retriever with a sentencetransformer ranker as follows : search = (retriever + ranker)
While trying to add documents to the pipeline (search.add(documents=documents), I got this error :

"""/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py in embedding(input, weight, padding_idx, max_norm, norm_type, scale_grad_by_freq, sparse)
2181 # remove once script supports set_grad_enabled
2182 no_grad_embedding_renorm(weight, input, max_norm, norm_type)
-> 2183 return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
2184
2185

IndexError: index out of range in self"""

the doc at https://raphaelsty.github.io/cherche/api/compose/Pipeline/
regarding the "call" method says:

If the batch_size_ranker, or batch_size_retriever it takes precedence over the batch_size. If the k_ranker, or k_retriever it takes precedence over the k parameter.

which is not really understandable, needs to be clarified (and could be interpreted as something misleading).

Regarding the k param, please note the following: if you define a retriever (say a tfidf one) with a k param of 20, followed by a ranker with a k param of 10, (your interested in top_k = 10 values at the end, but use 20 values at the retriever level) then a likely error one can make is to call the pipeline with a k value of 10. In this case indeed, it appears that the retriever uses a k value of 10.

Just curious if this project is still active. It looks great, thank for working on it!

https://raphaelsty.github.io/cherche/examples/eval_pipeline/ you say that one can find the explnaation about metrics here https://amitness.com/2020/08/information-retrieval-evaluation/ but it doesn't say what "precision" and "r-precision" are.

Create a retriever based on a sentence bert, passing a value, eg. 10, to k param.
It is not taken into account when calling the retriever (more values are returned)

    retriever = retrieve.Encoder(
        key='id',
        on=['content'],
        encoder=SentenceTransformer('sentence-transformers/all-MiniLM-L12-v2').encode,
        k = 10
    )
    retriever(documents=docs)

len(retriever(queries)[0]) > 10