DataLoader Class Reference

DataLoader allows to create a dataloader object to implement minibatch stochastic gradient descent. More...

#include <data_loader.h>

Public Member Functions
void	write_to_memory (dnnl::memory dst_mem_features, dnnl::memory dst_mem_labels)
	Method that writes the curr batch to memory and moves the index forward.
	DataLoader (std::string features_path, std::string labels_path, int _minibatch_size, std::vector< int > dataset_shape, dnnl::engine _eng)
	Construct a new Data Loader object. More...

Public Attributes
int	dataset_size
	Total number of samples.
int	minibatch_size
	Minibatch size.
std::vector< float >	dataset
	Vector containing the entire dataset.
std::vector< float >	dataset_labels
	Vector containing the labels.
std::vector< float >	curr_batch
	Vector containing the current batch that will be written to the engine.
std::vector< float >	curr_batch_labels
	Vector containing the labels that will be written to the engine.

Detailed Description

DataLoader allows to create a dataloader object to implement minibatch stochastic gradient descent.

Constructor & Destructor Documentation

DataLoader()

DataLoader::DataLoader	(	std::string	features_path,
		std::string	labels_path,
		int	_minibatch_size,
		std::vector< int >	dataset_shape,
		dnnl::engine	_eng
	)

Construct a new Data Loader object.

This allows the creation of a class which implements minibatch stochastic gradient descent

Parameters

features_path	path to the text file containing the flattened features (in row-major order)
labels_path	path to the labels corresponding to the features
_minibatch_size	size of the minibatch (-1 for full batch)
dataset_shape	shape of the single sample eg. {C} or {C, H, W}, the N will be stored in the dataset_size variable
_eng	oneAPI engine

---

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

• src/misc/include/data_loader.h