raft
raft copied to clipboard
[FEA] API to set random seed for IVF subsampling
In IVF-Flat and IVF-PQ, we generate random indices and shuffle or subsample the dataset using these indices before training. Currently a fixed seed is used to generate random indices.
This issue requests an option to control the random seed. PR #2077 had planned to add a parameter, but it was removed, because we were considerin alternative options (see discussion here).
Alternatively, we could consider random sampling seed as an implementation detail, and decide not to expose a user knob for it.