#include <TrainingData.h>

Inheritance diagram for panann::TrainingData:

Collaboration diagram for panann::TrainingData:

Public Types
enum	ScalingAlgorithm : uint8_t { ScalingAlgorithm::None = 0, ScalingAlgorithm::Simple, ScalingAlgorithm::StandardDeviation, ScalingAlgorithm::UniformNorm }

Public Member Functions
	TrainingData (const TrainingData &)=delete

TrainingData &	operator= (const TrainingData &)=delete

void	SetScalingAlgorithm (ScalingAlgorithm algorithm)

ScalingAlgorithm	GetScalingAlgorithm () const

void	SetSimpleScalingNewMin (double val)

double	GetSimpleScalingNewMin () const

void	SetSimpleScalingNewMax (double val)

double	GetSimpleScalingNewMax () const

void	SetStandardDeviationMultiplier (double val)

double	GetStandardDeviationMultiplier () const

void	SetUniformNormMultiplier (double val)

double	GetUniformNormMultiplier () const

void	Scale ()

void	Descale ()

void	ScaleInput (std::vector< double > *vec) const

void	ScaleOutput (std::vector< double > *vec) const

void	DescaleInput (std::vector< double > *vec) const

void	DescaleOutput (std::vector< double > *vec) const

void	FromSequentialData (const std::vector< double > &data, size_t input_length)

Protected Member Functions
void	CalculateMinMax ()

void	CalculateMean ()

void	CalculateStdev ()

void	CalculateUniformNorm ()

void	ScaleSimple ()

void	ScaleStandardDeviation ()

void	ScaleUniformNorm ()

void	DescaleSimple ()

void	DescaleStandardDeviation ()

void	DescaleUniformNorm ()

Detailed Description

The TrainingData is just a vector of Examples with some utility functions for scaling that data.

See also: Example

Member Enumeration Documentation

◆ ScalingAlgorithm

enum panann::TrainingData::ScalingAlgorithm : uint8_t

strong

Enumerator
None	Perform no scaling of the Examples.
Simple	Scale the Examples by calculating a simple factor which will shift the values in the Examples into the range [simple_scaling_new_min_, simple_scaling_new_max_] (which is [-1,1] by default). See also SetSimpleScalingNewMin SetSimpleScalingNewMax
StandardDeviation	Recenters the Examples around the mean and scales them by a factor of the standard deviation. See also SetStandardDeviationMultiplier
UniformNorm	Recenters the Examples around the mean and scales them by a factor of the uniform norm. See also SetUniformNormMultiplier

Member Function Documentation

◆ Descale()

void panann::TrainingData::Descale ( )

Descale the Examples in this TrainingData back to their original values before the data was scaled via Scale().
Uses the selected scaling algorithm.

See also: SetScalingAlgorithm; Scale

◆ DescaleInput()

void panann::TrainingData::DescaleInput ( std::vector< double > * vec ) const

Descale one vector of input.
Uses the scaling parameters calculated during a previous call to Scale.

See also: Scale

◆ DescaleOutput()

void panann::TrainingData::DescaleOutput ( std::vector< double > * vec ) const

Descale one vector of output.
Uses the scaling parameters calculated during a previous call to Scale.

See also: Scale

◆ FromSequentialData()

void panann::TrainingData::FromSequentialData	(	const std::vector< double > &	data,
		size_t	input_length
	)

Convert sequential data into examples.
Use this to create examples from time series data or other sets of sequential data.
Each example will have |input_length| input samples and one output sample - which will be the value from |data| immediately following those input samples.
We will create as many examples as possible from |data|.

Parameters

input_length	The number of input samples to put into each example.
data	An ordered set of samples. Must have at least input_length elements.

◆ Scale()

void panann::TrainingData::Scale ( )

Scale the Examples in this TrainingData via the selected scaling algorithm.
Use SetScalingAlgorithm to select a scaling algorithm.
The scaling parameters are calculated separately for input and output because these two sets may be in completely different scales.

See also: SetScalingAlgorithm; Descale

◆ ScaleInput()

void panann::TrainingData::ScaleInput ( std::vector< double > * vec ) const

Scale one vector of input.
Uses the scaling parameters calculated during a previous call to Scale.

See also: Scale

◆ ScaleOutput()

void panann::TrainingData::ScaleOutput ( std::vector< double > * vec ) const

Scale one vector of output.
Uses the scaling parameters calculated during a previous call to Scale.

See also: Scale

◆ SetScalingAlgorithm()

void panann::TrainingData::SetScalingAlgorithm ( ScalingAlgorithm algorithm )

Set the algorithm used to scale Examples in this TrainingData via Scale().
Default: ScalingAlgorithm::Simple

See also: ScalingAlgorithm; Scale

◆ SetSimpleScalingNewMax()

void panann::TrainingData::SetSimpleScalingNewMax ( double val )

Set the new maximum value used by the simple scaling algorithm.
When using simple scaling, Examples will be scaled to the range [simple_scaling_new_min_, simple_scaling_new_max_].
By default, this range is [-1, 1].

See also: SetSimpleScalingNewMin

◆ SetSimpleScalingNewMin()

void panann::TrainingData::SetSimpleScalingNewMin ( double val )

Set the new minimum value used by the simple scaling algorithm.
When using simple scaling, Examples will be scaled to the range [simple_scaling_new_min_, simple_scaling_new_max_].
By default, this range is [-1, 1].

See also: SetSimpleScalingNewMax

◆ SetStandardDeviationMultiplier()

void panann::TrainingData::SetStandardDeviationMultiplier ( double val )

Set the multiplier used by standard deviation scaling algorithm.
After recentering the Examples around their mean, we scale each value by this multiple of the standard deviation.
Default: 2.5

◆ SetUniformNormMultiplier()

void panann::TrainingData::SetUniformNormMultiplier ( double val )

Set the multiplier used by uniform norm scaling algorithm.
After recentering the Examples around their mean, we scale each value by this multiple of the uniform norm.
Default: 1.0

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

src/TrainingData.h
src/TrainingData.cc

Public Types

Public Member Functions

Protected Member Functions