|
|
|
|
|
|
by zX41ZdbW
970 days ago
|
|
|
I've implemented the same approach (non-parametric), and it worked well to speed up the search on my experiment with 384-dimensional embeddings of 30 million Hacker News comments. I presented it in my talk, where you can see the speed comparison: https://www.youtube.com/watch?v=hGRNcftpqAk |
|
|