Class TimeSeriesRegressionOptions<T>
Configuration options for time series regression models, which analyze data collected over time to identify patterns and make predictions.
public class TimeSeriesRegressionOptions<T> : RegressionOptions<T>Type Parameters
TThe numeric type used for calculations (typically double or float).
- Inheritance
-
TimeSeriesRegressionOptions<T>
- Derived
- Inherited Members
Remarks
Time series regression extends traditional regression analysis to account for the temporal nature of data, where observations are collected sequentially over time. These models can capture trends, seasonal patterns, and the effects of past values on current and future values.
For Beginners: Time series regression helps you analyze and predict data that changes over time, like stock prices, weather patterns, or monthly sales figures. Unlike regular regression that just looks for relationships between variables, time series regression also considers when things happened. It can detect patterns like "sales always increase in December" or "temperature today is related to temperature yesterday." This class lets you configure how the model analyzes these time-based patterns.
Properties
AutocorrelationCorrection
Gets or sets whether to apply autocorrelation correction to the model.
public bool AutocorrelationCorrection { get; set; }Property Value
- bool
True to apply autocorrelation correction, defaulting to true.
Remarks
Autocorrelation occurs when error terms in a time series are correlated across observations, violating standard regression assumptions. When this option is enabled, the model applies methods to correct for this correlation, improving the accuracy of coefficient estimates and predictions.
For Beginners: This determines whether the model should adjust for the fact that errors in time series data tend to be related to each other. With the default value of true, the model will make these adjustments, which usually improves accuracy. For example, if the model consistently underestimates values for several days in a row, this correction helps it recognize and fix that pattern. You should generally leave this enabled unless you have a specific reason to disable it.
IncludeTrend
Gets or sets whether to include a trend component in the model.
public bool IncludeTrend { get; set; }Property Value
- bool
True to include a trend component, defaulting to true.
Remarks
When enabled, the model will include a linear trend term to capture consistent upward or downward movement in the data. This helps the model account for long-term directional changes that aren't explained by other variables.
For Beginners: This determines whether the model should look for overall upward or downward movement in your data over time. With the default value of true, the model will try to identify if your data is generally increasing or decreasing (like a growing customer base or declining costs). If your data doesn't have a clear direction over time, you might set this to false to simplify the model.
LagOrder
Gets or sets the lag order, which determines how many previous time steps are used as predictors.
public int LagOrder { get; set; }Property Value
- int
The lag order, defaulting to 1.
Remarks
The lag order specifies how many previous observations are included as explanatory variables in the model. For example, a lag order of 2 means that values from both t-1 and t-2 time steps are used to predict the value at time t.
For Beginners: This controls how far back in time the model looks when making predictions. With the default value of 1, the model considers only the previous time period (like yesterday's value when predicting today's). If you set it to 3, the model would look at the three previous time periods (like the last three days' values). Higher values help capture longer-term patterns but require more data and can make the model more complex.
LossFunction
Gets or sets the loss function used for gradient computation and model training.
public ILossFunction<T>? LossFunction { get; set; }Property Value
- ILossFunction<T>
The loss function, defaulting to null (which will use MeanSquaredErrorLoss).
Remarks
The loss function determines how the model measures prediction errors during training. Different loss functions are appropriate for different types of problems and data characteristics. If null, the model will use Mean Squared Error (MSE) as the default loss function.
For Beginners: This determines how the model measures its mistakes during training. The default (null) uses Mean Squared Error, which works well for most time series forecasting tasks. You can provide a custom loss function if you have specific requirements, such as: - MeanAbsoluteErrorLoss for robustness to outliers - HuberLoss for a balance between MSE and MAE - Custom loss functions for domain-specific error metrics
ModelType
Gets or sets the specific type of time series model to use.
public TimeSeriesModelType ModelType { get; set; }Property Value
- TimeSeriesModelType
The time series model type, defaulting to ARIMA.
Remarks
Different time series model types have different capabilities and are suited to different types of data. ARIMA (AutoRegressive Integrated Moving Average) is a flexible and widely-used approach that can handle many common time series patterns.
For Beginners: This selects which specific algorithm the model uses to analyze your time series data. The default is ARIMA, which is a popular and versatile method that works well for many types of time data. Other options might include methods like exponential smoothing (good for data with clear trends and seasonality) or VAR (Vector AutoRegression, good for analyzing relationships between multiple time series). Unless you're familiar with the different methods, starting with ARIMA is usually a good choice.
SeasonalPeriod
Gets or sets the seasonal period of the time series data.
public int SeasonalPeriod { get; set; }Property Value
- int
The seasonal period, defaulting to 0 (no seasonality).
Remarks
The seasonal period represents how many time steps make up one complete cycle of seasonal pattern. For example, 12 for monthly data with yearly seasonality, 7 for daily data with weekly seasonality. A value of 0 indicates that no seasonal component should be included in the model.
For Beginners: This tells the model if your data has regular patterns that repeat at fixed intervals. The default value of 0 means "don't look for seasonal patterns." If you're analyzing monthly data and expect yearly patterns (like retail sales peaking every December), you would set this to 12. For daily data with weekly patterns, you'd use 7. For quarterly data with yearly patterns, you'd use 4. This helps the model recognize and predict these recurring cycles.