opennn::TrainingResults Struct Reference

Per-epoch error history and final summary produced by Optimizer::train(). More...

#include <optimizer.h>

Public Member Functions
	TrainingResults (const Index expected_epochs=0)
	Constructs a TrainingResults pre-sized for an expected epoch count.
virtual	~TrainingResults ()=default
	Virtual destructor.
string	write_stopping_condition () const
	Returns the canonical string name of the stopping condition.
float	get_training_error () const
	Final training error (last entry of training_error_history).
float	get_validation_error () const
	Final validation error (last entry of validation_error_history).
Index	get_epochs_number () const
	Number of epochs effectively run.
void	save (const filesystem::path &) const
	Saves the training results (history and final summary) to a file.
void	print (const string &message=string()) const
	Prints a human-readable summary to stdout.
Tensor< string, 2 >	write_override_results (const Index decimals=3) const
	Returns a 2D string table summarizing the training run.
void	resize_training_error_history (const Index)
	Resizes the training error history.
void	resize_validation_error_history (const Index)
	Resizes the validation error history.

Public Attributes
Optimizer::StoppingCondition	stopping_condition = Optimizer::StoppingCondition::None
	Stopping condition that ended the training run.
VectorR	training_error_history
	Per-epoch training error.
VectorR	validation_error_history
	Per-epoch validation error.
string	elapsed_time
	Total elapsed wall-clock time, formatted as "hh:mm:ss".
float	loss = NAN
	Final loss value.
Index	validation_failures = 0
	Number of consecutive validation-error increases at end of training.
float	loss_decrease = 0.0f
	Loss decrease observed in the last epoch.

Detailed Description

Per-epoch error history and final summary produced by Optimizer::train().

Constructor & Destructor Documentation

TrainingResults()

opennn::TrainingResults::TrainingResults

(

const Index

expected_epochs = 0

)

Constructs a TrainingResults pre-sized for an expected epoch count.

Parameters

expected_epochs

Initial capacity for the error history vectors.

~TrainingResults()

virtual opennn::TrainingResults::~TrainingResults ( )

virtualdefault

Virtual destructor.

Member Function Documentation

get_epochs_number()

Index opennn::TrainingResults::get_epochs_number

(

)

const

Number of epochs effectively run.

get_training_error()

float opennn::TrainingResults::get_training_error

(

)

const

Final training error (last entry of training_error_history).

get_validation_error()

float opennn::TrainingResults::get_validation_error

(

)

const

Final validation error (last entry of validation_error_history).

print()

void opennn::TrainingResults::print

(

const string &

message = string()

)

const

Prints a human-readable summary to stdout.

Parameters

message

Optional prefix message.

resize_training_error_history()

void opennn::TrainingResults::resize_training_error_history

(

const Index

)

Resizes the training error history.

Receives the new size.

resize_validation_error_history()

void opennn::TrainingResults::resize_validation_error_history

(

const Index

)

Resizes the validation error history.

Receives the new size.

save()

void opennn::TrainingResults::save

(

const filesystem::path &

)

const

Saves the training results (history and final summary) to a file.

Receives the destination path.

write_override_results()

Tensor< string, 2 > opennn::TrainingResults::write_override_results

(

const Index

decimals = 3

)

const

Returns a 2D string table summarizing the training run.

Parameters

decimals

Number of decimals used when formatting floats.

Returns

Tensor of (rows, columns) strings ready for pretty-printing.

write_stopping_condition()

string opennn::TrainingResults::write_stopping_condition

(

)

const

Returns the canonical string name of the stopping condition.

Returns

Name (e.g. "MaximumEpochsNumber").

Member Data Documentation

elapsed_time

string opennn::TrainingResults::elapsed_time

Total elapsed wall-clock time, formatted as "hh:mm:ss".

loss

float opennn::TrainingResults::loss = NAN

Final loss value.

loss_decrease

float opennn::TrainingResults::loss_decrease = 0.0f

Loss decrease observed in the last epoch.

stopping_condition

Optimizer::StoppingCondition opennn::TrainingResults::stopping_condition = Optimizer::StoppingCondition::None

Stopping condition that ended the training run.

training_error_history

VectorR opennn::TrainingResults::training_error_history

Per-epoch training error.

validation_error_history

VectorR opennn::TrainingResults::validation_error_history

Per-epoch validation error.

validation_failures

Index opennn::TrainingResults::validation_failures = 0

Number of consecutive validation-error increases at end of training.