kuilt-warp-ml/us.tractat.kuilt.warp/TrainingUpdate

TrainingUpdate

data class TrainingUpdate(val sampleCount: Long, val weights: List<Double>)

One peer's decoded training result: how many examples it trained on and the updated weight vector. Bridges a kernel run (or ReferenceTrainer.step) to FedAvg.

Constructors

TrainingUpdate

constructor(sampleCount: Long, weights: List<Double>)

Properties

sampleCount

val sampleCount: Long

weights

val weights: List<Double>

Functions

toContribution

fun toContribution(peer: ReplicaId, epoch: Long = 1): FedAvg

This update as a single-peer FedAvg contribution for peer at epoch.