This appears to be a hard dependency purely for KMeans used in initialization for HMMs. Pulling in sklearn as a top-level dependency brings in a large, non-JAX dependency chain for what's a single clustering call used at init time.
Can an HMM extra be added, and sklearn can be moved over to it?
This appears to be a hard dependency purely for KMeans used in initialization for HMMs. Pulling in sklearn as a top-level dependency brings in a large, non-JAX dependency chain for what's a single clustering call used at init time.
Can an HMM extra be added, and sklearn can be moved over to it?