DeepANTOptions<T>
Options & Configuration · AiDotNet.TimeSeries.AnomalyDetection
Options for the DeepAnT model. Defaults follow the paper where it states a value (32 filters; window 45 and
threshold 0.5, its A1 setting in Table 1) and otherwise the PyTorch implementation at github.com/datacubeR/DeepAnt;
see DeepANT.
Inheritance: ModelOptions → RegressionOptions<T> → TimeSeriesRegressionOptions<T> → DeepANTOptions<T>
Constructors
| Constructor | Summary |
|---|---|
DeepANTOptions() |
|
DeepANTOptions(DeepANTOptions<T> other) |
Properties
| Property | Summary |
|---|---|
double AnomalyThreshold { get; set; } |
The anomaly-score threshold in normalized units (the paper’s Table 1: 0.50 to 0.75; 0.5 for A1). |
double DropoutRate { get; set; } |
The dropout after the fully connected layer. |
double LearningRate { get; set; } |
The SGD learning rate. |
int BatchSize { get; set; } |
The windows per SGD step. |
int Epochs { get; set; } |
The training epochs. |
int Filters { get; set; } |
The filters of each convolution (the paper’s 32). |
int ForecastHorizon { get; set; } |
The prediction window p_w: the values predicted after each window (the paper predicts 1). |
int HiddenUnits { get; set; } |
The fully connected layer’s width. |
int KernelSize { get; set; } |
The width of each convolution. |
int PoolSize { get; set; } |
The width and stride of each max-pooling. |
int WindowSize { get; set; } |
The history window w: the values the predictor reads (the paper’s Table 1 uses 35 or 45). |