Setter for the embeddings property of VectorStore

[tibor-mach]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

The current implementations of vector databases override the embeddings property of langchain_core.vectorstores.VectorStore base class and use different names for the underlying object (the embedding function)

For example in Chroma this is called _embedding_function and in Milvus the same object is called embedding_func (not even private, so presumably users are expected to be able to rewrite it?).

This makes it difficult to work with these objects on a high level and swap one for another. One way to solve this would be to enforce the same name everywhere but a better solution is probably to add setters to these classes since it also allows users to set the Embeddings object later on, once the VectorStore object has already been initialised.

Motivation

It comes up when we are writing boilerplate code applied in different circumstances. Ideally, I would like to be able to write a class which instantiates a vector store based on given parameters independently of any specific embedding function (which I would like to set up later). and I would like to do it in a generic way without having to worry about the different names of the langchain_core.embeddings.Embeddings class object used by the vector store.

Proposal (If applicable)

I think that all that is needed is to add a setter to the embeddings property in each of the VectorStore implementations. For example in Chroma it looks like this and in this is the implementation in Milvus (where it is called embedding_func rather than _embedding_function).

The example below is for Chroma:

    @embedding.setter
    def embedding(self, new_value):
        if self._embedding_function:
            warnings.warn(
                """
                Embedding function was set already! Replace if you must but
                beware of using a different embedding function after having already
                embedded data with the previous one!
                """,
                UserWarning,
            )
        self._embedding_function = new_value

An alternative solution would be to allow users to use the Embeddings object as a callback in all VectorStore methods where it is used. I would actually prefer this as it fits my usecase better but it would be probably more work to do this.