optimizer.h File Reference

Declares the Optimizer abstract base class and the supporting EpochStats / OptimizerData / TrainingResults structures. More...

#include <functional>
#include "json.h"
#include "tensor_utilities.h"
#include "thread_safe_queue.h"

Go to the source code of this file.

Classes
struct	opennn::EpochStats
	Aggregate metrics produced for a single training or evaluation epoch. More...
class	opennn::Optimizer
	Abstract base class for every training algorithm in OpenNN. More...
struct	opennn::OptimizerData
	Per-optimizer scratch state shared across iterations. More...
struct	opennn::TrainingResults
	Per-epoch error history and final summary produced by Optimizer::train(). More...

Namespaces
namespace	opennn

Detailed Description

Declares the Optimizer abstract base class and the supporting EpochStats / OptimizerData / TrainingResults structures.

Concrete optimizers (StochasticGradientDescent, AdaptiveMomentEstimation, QuasiNewtonMethod, LevenbergMarquardtAlgorithm, ...) extend Optimizer to implement train(), reusing the base-class infrastructure for batch prefetching, stopping conditions and JSON I/O.