SimMTMOptions<T>
Options & Configuration · AiDotNet.Models.Options
Configuration options for SimMTM (Simple Pre-Training Framework for Masked Time-Series Modeling).
Inheritance: ModelOptions → RegressionOptions<T> → TimeSeriesRegressionOptions<T> → SimMTMOptions<T>
How It Works
SimMTM combines masked time series modeling with series-level similarity learning, recovering masked series by aggregating from similar unmasked series in the batch.
Reference: Dong et al., “SimMTM: A Simple Pre-Training Framework for Masked Time-Series Modeling”, NeurIPS 2023.
Constructors
| Constructor | Summary |
|---|---|
SimMTMOptions() |
Creates options with the default values. |
SimMTMOptions(SimMTMOptions<T> other) |
Creates a copy of another options instance. |
Properties
| Property | Summary |
|---|---|
double DropoutRate { get; set; } |
Dropout probability applied during training. |
double MaskRatio { get; set; } |
Fraction of time steps masked in each training series. |
double SimilarityTemperature { get; set; } |
Temperature of the series-wise similarity used to weight neighbors when reconstructing masked series. |
int ContextLength { get; set; } |
Number of past time steps the model reads as context. |
int ForecastHorizon { get; set; } |
Number of future time steps to forecast. |
int HiddenDimension { get; set; } |
Width of the encoder’s hidden representations. |
int NumHeads { get; set; } |
Number of attention heads. |
int NumLayers { get; set; } |
Number of encoder layers. |
int PatchLength { get; set; } |
Number of time steps in each patch the series is split into. |