BucketBatchSampler
Models & Types ยท AiDotNet.Data.Sampling
Groups sequences by length into buckets, then batches within each bucket to minimize padding.
Inheritance: DataSamplerBase โ BucketBatchSampler
Implements: IBatchSampler
How It Works
Bucket batching is essential for efficient NLP training. By grouping sequences of similar lengths together, padding waste is minimized and GPU utilization is maximized.
The sampler sorts samples by their length, divides them into buckets of roughly equal size, shuffles within each bucket, and yields batches from each bucket.
Constructors
| Constructor | Summary |
|---|---|
BucketBatchSampler(int[] sampleLengths, int batchSize, int numBuckets, bool dropLast, int? seed) |
Creates a new bucket batch sampler. |
Properties
| Property | Summary |
|---|---|
bool DropLast { get; set; } |
|
int BatchSize { get; set; } |
|
int Length { get; } |
Methods
| Method | Summary |
|---|---|
IEnumerable<int[]> GetBatchIndices() |