OpenSearchVectorSearch doesn't permit the user to specify a field name

[weaversam8]

Currently the OpenSearchVectorSearch class defaults to vector_field as the field name of the vector field in all vector similarity searches.

This works fine if you're populating your OpenSearch instance with data via LangChain, but doesn't work well if you're attempting to query a vector field with a different name that's been populated by some other process. For maximum utility, users should be able to customize which field is being queried.

[naveentatikonda]

@weaversam8 Thank you for your feedback. Will work on it and raise the PR soon to provide customized field names.

[weaversam8]

I'm actually working on a PR now, so I can give you a good starting point :)

I think the ability to specify the text field would also be helpful, so I'm planning on squeezing that into my PR as well.

[naveentatikonda]

Sounds good. Thank you for your contributions to OpenSearch.

[weaversam8]

Done, let me know what you think.

[naveentatikonda]

Done, let me know what you think.

@weaversam8 I just reviewed your PR, everything looks good except a couple of things:

1. We also need to update the script source function to use custom field name
2. What is the reason behind changing the space_type to “cosinesimil” instead of “cosineSimilarity” for “painless_scripting” ? Please refer to this k-NN documentation https://opensearch.org/docs/latest/search-plugins/knn/painless-functions/#function-types

I have addressed these changes in my PR. Also, I have included the changes to add custom vector_field and text_field while indexing.

[naveentatikonda]

@hwchase17 Could you please tag me and @vamshin if any new PRs or github issues are created that are related to OpenSearch. We would like to contribute by addressing those issues and reviewing those PRs.