Can I fed 500K documents in rank_bm25?

[ramsey-coding]

Thanks for this awesome library.

I am curious to know whether rank_bm25 can handle 500K documents. Each document has around 1000 words.

Looking forward to your feedback. I want to use the following functionality with rank_bm25:

from rank_bm25 import BM25Okapi

corpus = [
    "Hello there good man!",
    "It is quite windy in London",
    "How is the weather today?"
]

tokenized_corpus = [doc.split(" ") for doc in corpus]
bm25 = BM25Okapi(tokenized_corpus)


query = "windy London"
tokenized_query = query.split(" ")

doc_scores = bm25.get_scores(tokenized_query)
result = bm25.get_top_n(tokenized_query, corpus, n=1)

print(result)

[ramsey-coding]

@Witiko can you please provide any insight?

[Witiko]

@ramsey-coding I don't see a reason why it shouldn't. Have you tried?

[ramsey-coding]

@Witiko the problem is call to the bm25.get_top_n is very very slow :-(

It is taking ~5 second per call.

[ramsey-coding]

@dorianbrown the library is slow to retrieval from ~350K samples. Can you please guide what to do here?

[AmenRa]

Hi @ramsey-coding,

I have just released a new Python-based search engine called retriv.
It only takes ~40ms to query 8M documents on my machine.
If you try it, please, let me know if it works for your use case.

[nashid]

@AmenRa I am also interested in this feature. Would try out retriv.

[nocoolsandwich]

Better use elastichsearch.Python version can be slow makes you crazy

[AmenRa]

@nocoolsandwich

You should try my library retriv.
It takes 10 ms to search 10 million documents with BM25.