vectorStore.similaritySearch in javascript is not filtering

[dportabella]

function filter(metadata) {
  console.log(`filter. metadata: ${JSON.stringify(metadata)}`);
  return tags.some(tag => metadata.tags.includes(tag)) && (!category|| metadata.categories.includes(category));
}
const vectorStore = await FaissStore.loadFromPython(vectorStoreFile, embeddings);
const results = await vectorStore.similaritySearch(queryText, 3, {filter: filter});  

vectorStore.similaritySearch in javascript is not calling the filter function. how to solve it?

Note: The docs mentions "Filtering by metadata is currently not supported.", however, the source code seems to have it implemented.
https://js.langchain.com/v0.2/docs/integrations/vectorstores/faiss/

langchainjs/langchain-core/src/vectorstores.ts

Line 184 in 3cc45fe

similaritySearchWithScore(

Note: filtering by metadata works correctly in python:

def filter(metadata):
  return set(tags).intersection(metadata["tags"]) is not None and (category is None or category in metadata["categories"])
vector_store = FAISS.load_local(vector_store_file, embeddings, allow_dangerous_deserialization=True)
results = vector_store.similarity_search_with_score(query_text, k=3, filter=filter)

[gon-martinam]

Hi. I've also found myself in the same situation.
There's indeed no option for filtering by metadata in FaissStore class defined in Langchain JS

The method similaritySearchWithScore(...) you provided is defined in the interface VectorStoreInterface, so it needs to be implemented by any class that implements the interface.

If you check the implementation of the method similaritySearchWithScore(...) for FaissStore class, you'll be able to see that there's not the possibility to filter the results.

Something you can try is using a custom retriever that wraps the search functionality of FaissStore to implement the filter.