search

langchain-ai / langchain

🦜🔗 Build context-aware reasoning applications
https://python.langchain.com
MIT License
92.36k stars 14.76k forks source link

num_candidates as fetch_k won't work on langchain #25180

Open shadow66603 opened 1 month ago

shadow66603 commented 1 month ago

Checked other resources

Example Code

from langchain_elasticsearch import ElasticsearchStore, DenseVectorStrategy, BM25Strategy

elastic_vector_search = ElasticsearchStore( es_url=YOUR_URL, index_name=YOUR_INDEX, es_user=YOUR_LOGIN, es_params = {'verify_certs':False,'request_timeout':1000}, es_password=YOUR_PASSWORD, embedding=embeddings, strategy=DenseVectorStrategy() )

retriever = elastic_vector_search.as_retriever(search_type="similarity_score_threshold",search_kwargs={'score_threshold': 0.85, 'k':150,'fetch_k' : 10000}, include_original=True)

retriever.get_relevant_documents('query')

Error Message and Stack Trace (if applicable)

BadRequestError Traceback (most recent call last) Cell In[6], line 1 ----> 1 retriever.get_relevant_documents('réseau')

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core_api\deprecation.py:168, in deprecated..deprecate..warning_emitting_wrapper(*args, *kwargs) 166 warned = True 167 emit_warning() --> 168 return wrapped(args, **kwargs)

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core\retrievers.py:358, in BaseRetriever.get_relevant_documents(self, query, callbacks, tags, metadata, run_name, kwargs) 356 if run_name: 357 config["run_name"] = run_name --> 358 return self.invoke(query, config, kwargs)

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core\retrievers.py:219, in BaseRetriever.invoke(self, input, config, **kwargs) 217 except Exception as e: 218 run_manager.on_retriever_error(e) --> 219 raise e 220 else: 221 run_manager.on_retriever_end( 222 result, 223 )

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core\retrievers.py:212, in BaseRetriever.invoke(self, input, config, kwargs) 210 _kwargs = kwargs if self._expects_other_args else {} 211 if self._new_arg_supported: --> 212 result = self._get_relevant_documents( 213 input, run_manager=run_manager, _kwargs 214 ) 215 else: 216 result = self._get_relevant_documents(input, **_kwargs)

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core\vectorstores\base.py:1249, in VectorStoreRetriever._get_relevant_documents(self, query, run_manager) 1246 docs = self.vectorstore.similarity_search(query, self.search_kwargs) 1247 elif self.search_type == "similarity_score_threshold": 1248 docs_and_similarities = ( -> 1249 self.vectorstore.similarity_search_with_relevance_scores( 1250 query, self.searchkwargs 1251 ) 1252 ) 1253 docs = [doc for doc, in docs_and_similarities] 1254 elif self.search_type == "mmr":

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core\vectorstores\base.py:777, in VectorStore.similarity_search_with_relevance_scores(self, query, k, kwargs) 761 """Return docs and relevance scores in the range [0, 1]. 762 763 0 is dissimilar, 1 is most similar. (...) 773 List of Tuples of (doc, similarity_score). 774 """ 775 score_threshold = kwargs.pop("score_threshold", None) --> 777 docs_and_similarities = self._similarity_search_with_relevance_scores( 778 query, k=k, kwargs 779 ) 780 if any( 781 similarity < 0.0 or similarity > 1.0 782 for _, similarity in docs_and_similarities 783 ): 784 warnings.warn( 785 "Relevance scores must be between" 786 f" 0 and 1, got {docs_and_similarities}" 787 )

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_core\vectorstores\base.py:725, in VectorStore._similarity_search_with_relevance_scores(self, query, k, kwargs) 707 """ 708 Default similarity search with relevance scores. Modify if necessary 709 in subclass. (...) 722 List of Tuples of (doc, similarity_score) 723 """ 724 relevance_score_fn = self._select_relevance_score_fn() --> 725 docs_and_scores = self.similarity_search_with_score(query, k, kwargs) 726 return [(doc, relevance_score_fn(score)) for doc, score in docs_and_scores]

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\langchain_elasticsearch\vectorstores.py:883, in ElasticsearchStore.similarity_search_with_score(self, query, k, filter, custom_query, doc_builder, **kwargs) 877 if ( 878 isinstance(self._store.retrieval_strategy, DenseVectorStrategy) 879 and self._store.retrieval_strategy.hybrid 880 ): 881 raise ValueError("scores are currently not supported in hybrid mode") --> 883 hits = self._store.search( 884 query=query, k=k, filter=filter, custom_query=custom_query 885 ) 886 return _hits_to_docs_scores( 887 hits=hits, 888 content_field=self.query_field, 889 doc_builder=doc_builder, 890 )

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\elasticsearch\helpers\vectorstore_sync\vectorstore.py:274, in VectorStore.search(self, query, query_vector, k, num_candidates, fields, filter, custom_query) 271 query_body = custom_query(query_body, query) 272 logger.debug(f"Calling custom_query, Query body now: {query_body}") --> 274 response = self.client.search( 275 index=self.index, 276 **query_body, 277 size=k, 278 source=True, 279 source_includes=fields, 280 ) 281 hits: List[Dict[str, Any]] = response["hits"]["hits"] 283 return hits

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\elasticsearch_sync\client\utils.py:446, in _rewrite_parameters..wrapper..wrapped(*args, *kwargs) 443 except KeyError: 444 pass --> 446 return api(args, **kwargs)

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\elasticsearch_sync\client__init.py:4119, in Elasticsearch.search(self, index, aggregations, aggs, allow_no_indices, allow_partial_search_results, analyze_wildcard, analyzer, batched_reduce_size, ccs_minimize_roundtrips, collapse, default_operator, df, docvalue_fields, error_trace, expand_wildcards, explain, ext, fields, filter_path, force_syntheticsource, from, highlight, human, ignore_throttled, ignore_unavailable, indices_boost, knn, lenient, max_concurrent_shard_requests, min_compatible_shard_node, min_score, pit, post_filter, pre_filter_shard_size, preference, pretty, profile, q, query, rank, request_cache, rescore, rest_total_hits_as_int, retriever, routing, runtime_mappings, script_fields, scroll, search_after, search_type, seq_no_primary_term, size, slice, sort, source, source_excludes, source_includes, stats, stored_fields, suggest, suggest_field, suggest_mode, suggest_size, suggest_text, terminate_after, timeout, track_scores, track_total_hits, typed_keys, version, body) 4117 if body is not None: 4118 __headers["content-type"] = "application/json" -> 4119 return self.perform_request( # type: ignore[return-value] 4120 "POST", 4121 path, 4122 params=query, 4123 headers=headers, 4124 body=body, 4125 endpoint_id="search", 4126 path_parts=__path_parts, 4127 )

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\elasticsearch_sync\client_base.py:271, in BaseClient.perform_request(self, method, path, params, headers, body, endpoint_id, path_parts) 255 def perform_request( 256 self, 257 method: str, (...) 264 path_parts: Optional[Mapping[str, Any]] = None, 265 ) -> ApiResponse[Any]: 266 with self._otel.span( 267 method, 268 endpoint_id=endpoint_id, 269 path_parts=path_parts or {}, 270 ) as otel_span: --> 271 response = self._perform_request( 272 method, 273 path, 274 params=params, 275 headers=headers, 276 body=body, 277 otel_span=otel_span, 278 ) 279 otel_span.set_elastic_cloud_metadata(response.meta.headers) 280 return response

File ~\AppData\Local\Programs\Python\Python39\lib\site-packages\elasticsearch_sync\client_base.py:352, in BaseClient._perform_request(self, method, path, params, headers, body, otel_span) 349 except (ValueError, KeyError, TypeError): 350 pass --> 352 raise HTTP_EXCEPTIONS.get(meta.status, ApiError)( 353 message=message, meta=meta, body=resp_body 354 ) 356 # 'X-Elastic-Product: Elasticsearch' should be on every 2XX response. 357 if not self._verified_elasticsearch: 358 # If the header is set we mark the server as verified.

BadRequestError: BadRequestError(400, 'illegal_argument_exception', '[num_candidates] cannot be less than [k]')

Description

I am trying to fetch 150 documents (k) and I specify the fetch_k (supposed to be num_candidates) and it does not work: '[num_candidates] cannot be less than [k]')

Two options:

System Info

langchain-0.2.12 python 3.9

venkatram-dev commented 3 weeks ago

Great catch. So technically it is correct to have 'k':150,'fetch_k' : 10000

https://github.com/langchain-ai/langchain/blob/master/libs/community/langchain_community/vectorstores/elasticsearch.py#L636C1-L637C87

        k: Number of Documents to return. Defaults to 4.
        fetch_k (int): Initial Number of Documents to fetch and to pass to knn num_candidates.

But If we try to look at the error  '[num_candidates] cannot be less than [k]'),
what could be happening is that from elastic search the initial fetch itself is returning only about 100 docs (less than 150).
From the 100 docs, its not able to return 150 docs and hence the errors.

I would suggest to try lesser number for k and fetch_k initially and try few combinations.

'k':1,'fetch_k' : 10
....
'k':0,'fetch_k' : 100
....
....
'k':100,'fetch_k' : 100
    ....
....
'k':100,'fetch_k' : 100

It might be a dormant bug, You might also want to try the latest lanchain version and see if it got fixed.