TiDEOptions<T>

Options & Configuration · AiDotNet.Models.Options

Options for TiDEModel — Time-series Dense Encoder (Das et al., “Long-term Forecasting with TiDE: Time-series Dense Encoder”, TMLR 2023). A pure-MLP encoder/decoder with a linear residual; on long-horizon benchmarks it matches or beats transformers at a fraction of the cost. Faithful core: a ReLU encoder MLP + decoder projection + a linear skip from the input window.

Inheritance: ModelOptions → RegressionOptions<T> → TimeSeriesRegressionOptions<T> → TiDEOptions<T>

For Beginners

this model squeezes the recent history through a small neural network (“encoder”) to a compact summary, expands that summary into the forecast (“decoder”), and also adds a simple straight-line shortcut from the input so it never does worse than a linear model. The HiddenSize knob sets how big that compact summary is.

How It Works

TiDE encodes the lookback window with a multi-layer-perceptron (MLP) encoder into a latent vector of size HiddenSize, decodes that to the forecast horizon, and adds a linear residual mapped directly from the input window. Being an MLP, it captures non-linear structure that the purely-linear DLinear/NLinear cannot, while remaining far cheaper than attention-based models.

Typical sizes: LookbackWindow 24–96, HiddenSize 32–256 (64 is a good default), ForecastHorizon 1 for the supervised harness.

Reference: Abhimanyu Das, Weihao Kong, Andrew Leach, Shaan Mathur, Rajat Sen, Rose Yu. “Long-term Forecasting with TiDE: Time-series Dense Encoder.” TMLR 2023. https://arxiv.org/abs/2304.08424

Constructors

Constructor Summary
TiDEOptions() Creates a new TiDEOptions with default values.
TiDEOptions(TiDEOptions<T> other) Creates a deep copy of an existing TiDEOptions.

Properties

Property Summary
double LearningRate { get; set; } Step size for gradient-descent optimization.
int BatchSize { get; set; } Number of training samples processed per gradient update.
int Epochs { get; set; } Number of full passes over the training data.
int ForecastHorizon { get; set; } Forecast horizon.
int HiddenSize { get; set; } Latent dimension of the MLP encoder (and decoder hidden width).
int LookbackWindow { get; set; } History length the encoder sees (input window).