Support/test multiple Vector DBs #56
Description
Description
Validate that all our supported Vector Databases work as expected. At this point in time, we have only used HNSWLib.
Impact
- Who: Developers
- What: Support connecting to different DBs and test it out.
- Why: Make vCache more usable as organizations require to use specific DBs.
Proposed Solution
We should verify our solution works with more than one database, and if required, build a class that the users can instantiate as a wildcard DB that automatically connects with the existing DB.
Acceptance Criteria
- Followed coding conventions
- Implemented tests for different Vector DBs
- Fix this: FAISS reset loses dimension information by setting index = None (VectorDB and Judge Parallelism #55 (comment))
- Fix this: Potential bug: filtering IDs after transforming similarity can cause an index mismatch. Filter both arrays together or use enumerate (VectorDB and Judge Parallelism #55 (comment))
Risks & Dependencies
Limited by the current architecture of vCache. It might require an abstract class that users can use, and implicitly cast to the right database automatically.
Additional Context
No extra context.