Skip to content

Too large minhashLSH index #207

Description

@bryanyzhu

Hi, I have a question about large-scale LSH index. If I have billions of documents, I suppose even 1T RAM is not enough to do in-memory LSH, is there any recommended way to use datasketch for this scenario? Thank you.

I also opened an issue #206 because for a small subset on my local machine (a 6GB pickle file containing pre-computed minhashes), if I use LSH threhold of 0.5 for inserting to minhashLSH, it takes 31GB RAM. If I use leanminhash, it takes 26GB. Then I can do a simple extrapolation, for 600GB pre-computed minhashes, the indexing process will take 3T RAM. This is just too much. Maybe mapping it to disk could be a viable solution. Looking forward to any suggestions, thank you.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions