I am wondering whether or where I can find the sample code for this blog Speeding up merging of HNSW graphs - Elasticsearch Labs?
Hi @Chenzhe_Jin,
Welcome! To confirm are you looking for a notebook containing the code for the experiment rather than the discussed pseudocode? Normally search labs accompanying notebooks are located in the elasticsearch-labs GitHub repo under the supporting-blog-content folder, but I expect there isn't one for this piece.
The author @Tom_Veasey and @mayya may be able to help.
This change is incorporated in Lucene (a library that Elasticsearch uses underneath).
You can study the code in the following PR.
Thanks for your reply! I am also wondering if there is some code or instructions we can follow to reproduce the experiments in the blog and PR?
Hi, I have another question when I read the code. Can your merge strategy support parallelism? When I read the code, I find there is no parallelism design. Thanks for your patience.
No, currently there is no parallelism.
We go through all segment graphs and add them incrementally to the biggest graph.