Deep::StreamAlignedBatcher Class Reference

Builds batches with a fixed, grouped number of examples per class – see the file-level note for why this exists and what it deliberately doesn't own. More...

#include <StreamAlignedBatcher.h>

Public Member Functions
	StreamAlignedBatcher (const float *X, const float *Y, const int *labels, size_t numSamples, size_t xStride, size_t yStride, int numClasses, int perClass, unsigned seed=42)
size_t	NumBatchesPerEpoch () const noexcept
	Number of complete batches obtainable per epoch, limited by the rarest class's sample count (matches the Python version's behavior – some samples from more common classes go unused each epoch, rather than padding/repeating to force even usage).
void	GetBatch (float *X_out, float *Y_out, int *labels_out) noexcept
	Fills the next batch into caller-provided, pre-allocated output buffers. Rows are GROUPED by class: rows [0, perClass) are class 0, [perClass, 2*perClass) are class 1, etc. – same convention as the Python version, so per-class slicing for cache-averaging is a trivial contiguous range.
int	GetBatchSize () const noexcept
	Rows per batch: numClasses * perClass.
int	GetPerClass () const noexcept
	Examples per class in each batch, as given at construction.
int	GetNumClasses () const noexcept
	Number of distinct classes, as given at construction.
size_t	GetXStride () const noexcept
	Floats per X row, as given at construction.
size_t	GetYStride () const noexcept
	Floats per Y row, as given at construction.

Detailed Description

Builds batches with a fixed, grouped number of examples per class – see the file-level note for why this exists and what it deliberately doesn't own.

Constructor & Destructor Documentation

StreamAlignedBatcher()

Deep::StreamAlignedBatcher::StreamAlignedBatcher	(	const float *	X,
		const float *	Y,
		const int *	labels,
		size_t	numSamples,
		size_t	xStride,
		size_t	yStride,
		int	numClasses,
		int	perClass,
		unsigned	seed = 42 )

Parameters

X	Flat, row-major input data, numSamples rows of xStride floats each. NOT owned.
Y	Flat, row-major target data, numSamples rows of yStride floats each. NOT owned.
labels	Integer class label per sample, length numSamples. NOT owned.
numSamples	Total number of samples in X/Y/labels.
xStride	Floats per X row.
yStride	Floats per Y row.
numClasses	Number of distinct classes (labels expected in [0, numClasses)).
perClass	Examples per class per batch (batch size = numClasses * perClass).
seed	RNG seed for per-class shuffling; defaults to a fixed value for reproducible batch order run-to-run.

Member Function Documentation

GetBatch()

void Deep::StreamAlignedBatcher::GetBatch ( float * X_out, float * Y_out, int * labels_out )

noexcept

Fills the next batch into caller-provided, pre-allocated output buffers. Rows are GROUPED by class: rows [0, perClass) are class 0, [perClass, 2*perClass) are class 1, etc. – same convention as the Python version, so per-class slicing for cache-averaging is a trivial contiguous range.

Parameters

X_out	Must be sized (numClasses*perClass) * xStride.
Y_out	Must be sized (numClasses*perClass) * yStride.
labels_out	Must be sized (numClasses*perClass).

GetBatchSize()

int Deep::StreamAlignedBatcher::GetBatchSize ( ) const

inlinenoexcept

Rows per batch: numClasses * perClass.

GetNumClasses()

int Deep::StreamAlignedBatcher::GetNumClasses ( ) const

inlinenoexcept

Number of distinct classes, as given at construction.

GetPerClass()

int Deep::StreamAlignedBatcher::GetPerClass ( ) const

inlinenoexcept

Examples per class in each batch, as given at construction.

GetXStride()

size_t Deep::StreamAlignedBatcher::GetXStride ( ) const

inlinenoexcept

Floats per X row, as given at construction.

GetYStride()

size_t Deep::StreamAlignedBatcher::GetYStride ( ) const

inlinenoexcept

Floats per Y row, as given at construction.

NumBatchesPerEpoch()

size_t Deep::StreamAlignedBatcher::NumBatchesPerEpoch ( ) const

noexcept

Number of complete batches obtainable per epoch, limited by the rarest class's sample count (matches the Python version's behavior – some samples from more common classes go unused each epoch, rather than padding/repeating to force even usage).

---

The documentation for this class was generated from the following files:

• include/deepity/utils/StreamAlignedBatcher.h
• src/utils/StreamAlignedBatcher.cpp