ElasticDistributedSampler
Models & Types ยท AiDotNet.Data.Sampling
Distributed sampler that evenly divides data across multiple workers with elastic scaling support.
Inheritance: DataSamplerBase โ ElasticDistributedSampler
How It Works
Ensures each distributed worker sees a non-overlapping subset of the data each epoch. Supports dynamic resizing when workers join or leave (elastic training). Each worker gets dataset_size / num_replicas samples per epoch.
Constructors
| Constructor | Summary |
|---|---|
ElasticDistributedSampler(ElasticDistributedSamplerOptions options) |
Creates a new elastic distributed sampler. |
Properties
| Property | Summary |
|---|---|
int DatasetSize { get; } |
Gets the total dataset size. |
int Length { get; } |
|
int NumReplicas { get; } |
Gets the current number of replicas. |
int Rank { get; } |
Gets the current rank. |
Methods
| Method | Summary |
|---|---|
void Rescale(int numReplicas, int rank) |
Dynamically updates the number of replicas and rank for elastic scaling. |