Class ExponentialSmoothingOptions<T>
Configuration options for Exponential Smoothing, a time series forecasting method that gives exponentially decreasing weights to older observations.
public class ExponentialSmoothingOptions<T> : TimeSeriesRegressionOptions<T>Type Parameters
TThe numeric type used for calculations (typically double or float).
- Inheritance
-
ExponentialSmoothingOptions<T>
- Inherited Members
Remarks
Exponential smoothing is a popular forecasting method for time series data that works by applying weighted averages where the weights decrease exponentially as observations get older. This means recent observations have more influence on forecasts than older observations.
For Beginners: Exponential smoothing is like a weighted average that gives more importance to recent data points and less importance to older ones. Imagine predicting tomorrow's temperature: you might care more about today's temperature than what happened two weeks ago. This method works similarly, gradually "forgetting" older data while focusing on newer trends. It's particularly good for data that changes over time and where recent observations are more relevant for prediction. This class lets you configure how quickly the model "forgets" old data and how it handles trends and seasonal patterns.
Properties
GridSearchStep
Gets or sets the step size for grid search when optimizing smoothing parameters.
public double GridSearchStep { get; set; }Property Value
- double
The grid search step size, defaulting to 0.1.
Remarks
When the model is being trained, it can search for optimal values of alpha, beta, and gamma by testing different combinations. This parameter controls how finely the search grid is divided, with smaller values providing more precise but more computationally expensive optimization.
For Beginners: This controls how thoroughly the model searches for the best smoothing parameters. With the default value of 0.1, the model will try values like 0.1, 0.2, 0.3, etc. A smaller value like 0.01 would make it try more values (0.01, 0.02, 0.03, etc.), potentially finding better parameters but taking longer to train. Think of it as the precision of the model's search for optimal settings - smaller steps mean a more detailed search but require more computation time.
InitialAlpha
Gets or sets the initial smoothing factor (alpha) that controls the weight given to recent observations.
public double InitialAlpha { get; set; }Property Value
- double
The initial alpha value, defaulting to 0.3.
Remarks
Alpha is the primary smoothing parameter that determines how quickly the influence of older observations decreases. Values range from 0 to 1, with higher values giving more weight to recent observations and less weight to older ones.
For Beginners: This controls how quickly the model "forgets" older data points. With the default value of 0.3, the model gives moderate importance to recent observations while still considering older data. A higher value (closer to 1) would make the model focus almost entirely on the most recent data points, making it react quickly to changes but potentially overreacting to random fluctuations. A lower value (closer to 0) would make the model consider a longer history, resulting in smoother forecasts that might be slower to detect new trends. Think of it as adjusting how "short-term focused" your predictions should be.
InitialBeta
Gets or sets the initial trend smoothing factor (beta) that controls the weight given to the trend component.
public double InitialBeta { get; set; }Property Value
- double
The initial beta value, defaulting to 0.1.
Remarks
Beta is the trend smoothing parameter that determines how the model responds to changes in the trend of the time series. Values range from 0 to 1, with higher values making the trend component more responsive to recent changes in the trend.
For Beginners: This controls how quickly the model adapts to changes in the direction of your data. With the default value of 0.1, the model gradually adjusts its understanding of trends. For example, if your sales data has been increasing by 5% each month but suddenly starts increasing by 10% each month, a higher beta value would make the model recognize and adapt to this change in trend more quickly. Lower values create more stable trend estimates that don't change dramatically with short-term fluctuations. This is only used when UseTrend is set to true.
InitialGamma
Gets or sets the initial seasonal smoothing factor (gamma) that controls the weight given to the seasonal component.
public double InitialGamma { get; set; }Property Value
- double
The initial gamma value, defaulting to 0.1.
Remarks
Gamma is the seasonal smoothing parameter that determines how the model responds to changes in the seasonal pattern of the time series. Values range from 0 to 1, with higher values making the seasonal component more responsive to recent changes in the seasonal pattern.
For Beginners: This controls how quickly the model adapts to changes in seasonal patterns in your data. With the default value of 0.1, the model gradually updates its understanding of seasonal effects. For example, if your retail business always had a December sales spike of 50% but this year it's 70%, a higher gamma value would make the model adjust its seasonal expectations more quickly. Lower values create more stable seasonal estimates that don't change dramatically with one unusual season. This is only used when UseSeasonal is set to true.
UseSeasonal
Gets or sets whether to include a seasonal component in the exponential smoothing model.
public bool UseSeasonal { get; set; }Property Value
- bool
True to include a seasonal component, defaulting to false.
Remarks
When enabled, the model will account for seasonal patterns in the data (regular fluctuations that repeat at fixed intervals). This corresponds to using Holt-Winters' Method when both trend and seasonality are enabled.
For Beginners: This determines whether the model should look for and predict repeating patterns in your data. With the default value of false, the model ignores potential seasonal effects. If your data has regular patterns that repeat (like higher sales every December, or higher website traffic every weekend), you should set this to true. You'll also need to specify the seasonal period in the base SeasonalPeriod property (like 12 for monthly data with yearly patterns, or 7 for daily data with weekly patterns).
UseTrend
Gets or sets whether to include a trend component in the exponential smoothing model.
public bool UseTrend { get; set; }Property Value
- bool
True to include a trend component, defaulting to true.
Remarks
When enabled, the model will account for trends in the data (consistent upward or downward movement). This corresponds to using either Simple Exponential Smoothing (when false) or Holt's Linear Method (when true).
For Beginners: This determines whether the model should look for and predict continuing upward or downward movement in your data. With the default value of true, the model will identify trends (like steadily increasing sales or gradually decreasing costs) and use them in predictions. If your data doesn't show clear directional movement over time, or if you want simpler predictions that don't assume trends will continue, you might set this to false.