forecast

Documentation

Overview

Package forecast provides an interface for custom forecasting algorithms.

Index

• Variables
• func ConfidenceLevelToZ(level float64) float64
• type Confidence
  • func Forecast(ctx context.Context, sdf interface{}, r *dataframe.Range, ...) (interface{}, []Confidence, float64, error)
• type ConfidenceInterval
  • func DriftConfidenceInterval(pred, level, sigmaHat float64, T, h uint) ConfidenceInterval
  • func MeanConfidenceInterval(pred, level, sigmaHat float64, T uint) ConfidenceInterval
  • func NaïveConfidenceInterval(pred, level, sigmaHat float64, h uint) ConfidenceInterval
  • func SeasonalNaïveConfidenceInterval(pred, level, sigmaHat float64, h, seasonalPeriod uint) ConfidenceInterval
  • func (c ConfidenceInterval) NormalError() float64
  • func (c ConfidenceInterval) String() string
• type EvaluationFunc
• type EvaluationFuncOptions
• type ForecastingAlgorithm

Variables

var ErrIndeterminate = errors.New("indeterminate")

ErrIndeterminate indicates that the result of a calculation is indeterminate.

var ErrInsufficientDataPoints = errors.New("insufficient data points or nil values found")

ErrInsufficientDataPoints signifies that a particular forecasting algorithm requires more data points to operate.

var ErrMismatchLen = errors.New("mismatch length")

ErrMismatchLen signifies that there is a mismatch between the length of 2 Series or slices.

Functions

func ConfidenceLevelToZ

func ConfidenceLevelToZ(level float64) float64

ConfidenceLevelToZ returns the Z value for a given confidence level. level must be between 0 and 1 (exclusive).

level: 0.75 (75%) => 1.15 (approx)

level: 0.95 (95%) => 1.96 (approx)

level: 0.99 (99%) => 2.58 (approx)

See: https://otexts.com/fpp2/prediction-intervals.html

Types

type Confidence

type Confidence map[float64]ConfidenceInterval

Confidence contains the confidence intervals for various confidence levels. The key must be between 0 and 1 (exclusive). A confidence level of 95% is represented by 0.95.

func Forecast

func Forecast(ctx context.Context, sdf interface{}, r *dataframe.Range, alg ForecastingAlgorithm, cfg interface{}, n uint, evalFunc EvaluationFunc) (interface{}, []Confidence, float64, error)

Forecast predicts the next n values of sdf using the forecasting algorithm alg. cfg is required to configure the parameters of the algorithm. r is used to select a subset of sdf to be the "training set". Values after r form the "validation set". evalFunc can be set to measure the quality of the predictions. sdf can be a SeriesFloat64 or a DataFrame. DataFrame input is not yet implemented.

NOTE: You can find basic forecasting algorithms in forecast/algs subpackage.

type ConfidenceInterval

type ConfidenceInterval struct {
	// Upper bounds
	Upper float64

	// Lower bounds
	Lower float64

	// Normal can be set to signify that errors are normally distributed with a mean of zero.
	Normal bool
}

ConfidenceInterval represents an estimated range of values that includes the forecasted value within its bounds.

func DriftConfidenceInterval

func DriftConfidenceInterval(pred, level, sigmaHat float64, T, h uint) ConfidenceInterval

DriftConfidenceInterval - see https://otexts.com/fpp2/prediction-intervals.html

func MeanConfidenceInterval

func MeanConfidenceInterval(pred, level, sigmaHat float64, T uint) ConfidenceInterval

MeanConfidenceInterval - see https://otexts.com/fpp2/prediction-intervals.html

func NaïveConfidenceInterval

func NaïveConfidenceInterval(pred, level, sigmaHat float64, h uint) ConfidenceInterval

NaïveConfidenceInterval - see https://otexts.com/fpp2/prediction-intervals.html

func SeasonalNaïveConfidenceInterval

func SeasonalNaïveConfidenceInterval(pred, level, sigmaHat float64, h, seasonalPeriod uint) ConfidenceInterval

SeasonalNaïveConfidenceInterval - see https://otexts.com/fpp2/prediction-intervals.html

func (ConfidenceInterval) NormalError

func (c ConfidenceInterval) NormalError() float64

NormalError returns the error, assuming it is normally distributed with a mean of zero.

func (ConfidenceInterval) String

func (c ConfidenceInterval) String() string

String implements fmt.Stringer interface.

type EvaluationFunc

type EvaluationFunc func(ctx context.Context, validationSet, forecastSet []float64, opts *EvaluationFuncOptions) (float64, int, error)

EvaluationFunc compares the validationSet and forecastSet and calculates the error. See the validation subpackage for various approaches to calculating the error.

type EvaluationFuncOptions

type EvaluationFuncOptions struct {

	// SkipInvalids will skip Inf and NaN values.
	// If set to false (default) and an invalid value is encountered, then an ErrIndeterminate is returned.
	SkipInvalids bool
}

EvaluationFuncOptions is used to modify the behavior of the EvaluationFunc.

type ForecastingAlgorithm

type ForecastingAlgorithm interface {

	// Configure sets the various parameters for the algorithm.
	// config must be a struct that the particular algorithm recognizes.
	Configure(config interface{}) error

	// Load loads historical data.
	// Some forecasting algorithms do not tolerate nil values and will require interpolation.
	// r is used to limit which rows of sf are loaded. Prediction will always begin
	// from the row after that defined by r. r can be thought of as defining a "training set".
	Load(ctx context.Context, sf *dataframe.SeriesFloat64, r *dataframe.Range) error

	// Predict forecasts the next n values for the loaded data.
	//
	// NOTE: Not all forecasting algorithms return the confidence values.
	Predict(ctx context.Context, n uint) (*dataframe.SeriesFloat64, []Confidence, error)

	// Evaluate will measure the quality of the predicted values based on the evaluation calculation defined by evalFunc.
	// It will compare the error between sf and the values from the end of the loaded data ("validation set").
	// sf is usually the output of the Predict method.
	//
	// NOTE: You can use the functions directly from the validation subpackage if you need to do something
	// other than that described above.
	Evaluate(ctx context.Context, sf *dataframe.SeriesFloat64, evalFunc EvaluationFunc) (float64, error)
}

ForecastingAlgorithm defines the methods that all forecasting algorithms must implement.