Skip to content

Conversation

@bjchambers
Copy link
Contributor

@bjchambers bjchambers commented Jun 3, 2024

This required denormalizing the text_embedding of target nodes into
the edge, making it easier to guide traversal based on distance to a
query.

Also fixed a bug in keyword linking which didn't create links to old
nodes with the given keyword.

@bjchambers bjchambers requested review from cbornet and kerinin June 3, 2024 20:14
@bjchambers bjchambers changed the title draft: feat: Dynamic, MMR-based traversal feat: Dynamic, MMR-based traversal Jun 4, 2024
@bjchambers bjchambers marked this pull request as ready for review June 4, 2024 19:17
@bjchambers
Copy link
Contributor Author

@kerinin @cbornet this should be ready for review. I left two comments about things that may be worth doing in this iteration (vectorized cosine-similarity computation and score thresholds). Otherwise, it is tested and working. Things that may be especially worth looking at:

  • Naming / logic in the loops (extractors and mmr traversal). This was somewhat tricky, and I'm sure we can improve readability. I'll revisit when I have time too.
  • Changes to edge extractors to denormalize the embeddings into the edges.
This required denormalizing the `text_embedding` of target nodes into the edge, making it easier to guide traversal based on distance to a query. Also fixed a bug in keyword linking which didn't create links to old nodes with the given keyword.
@nicoloboschi nicoloboschi changed the title feat: Dynamic, MMR-based traversal [knowledge-store] feat: Dynamic, MMR-based traversal Jun 7, 2024
@bjchambers bjchambers merged commit 2cfedc4 into main Jun 7, 2024
@bjchambers bjchambers deleted the mmr-traversal branch June 7, 2024 17:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

5 participants