DCRNNOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for DCRNN (Diffusion Convolutional Recurrent Neural Network).

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

For Beginners

DCRNN was specifically designed for traffic forecasting by combining two powerful ideas:

The Key Insight: Traffic flow on road networks can be modeled as a diffusion process - like how congestion spreads through a network. DCRNN captures this with diffusion convolution while using an encoder-decoder architecture for multi-step prediction.

What Problems Does DCRNN Solve?

  • Traffic speed/flow prediction on road networks
  • Air quality forecasting across sensor networks
  • Subway ridership prediction
  • Any spatial-temporal forecasting where diffusion dynamics matter

How DCRNN Works:

  1. Diffusion Convolution: Models spatial dependencies as bidirectional random walks
  2. Diffusion GRU: Replaces matrix multiplications in GRU with diffusion convolution
  3. Encoder-Decoder: Encoder captures history, decoder generates predictions
  4. Scheduled Sampling: Gradually transitions from ground truth to predictions during training

DCRNN Architecture:

  • Encoder: Stacked DCGRU layers that encode input sequence
  • Decoder: Stacked DCGRU layers that generate output sequence
  • Diffusion: D_O^(K) + D_I^(K) bidirectional diffusion matrices
  • Output: Linear projection to forecast values

Key Benefits:

  • Captures spatial dependencies through diffusion process (not just adjacency)
  • Multi-step prediction through encoder-decoder architecture
  • Scheduled sampling prevents exposure bias during training
  • Bidirectional diffusion captures both upstream and downstream effects

How It Works

DCRNN combines diffusion convolution with sequence-to-sequence architecture for spatial-temporal forecasting on graph-structured data.

Reference: Li et al., “Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting”, ICLR 2018. https://arxiv.org/abs/1707.01926

Constructors

Constructor Summary
DCRNNOptions() Initializes a new instance of the DCRNNOptions class with default values.
DCRNNOptions(DCRNNOptions<T> other) Initializes a new instance by copying from another instance.

Properties

Property Summary
bool UseScheduledSampling { get; set; } Gets or sets whether to use scheduled sampling during training.
double DropoutRate { get; set; } Gets or sets the dropout rate for regularization.
double LearningRate { get; set; } Gets or sets the initial Adam learning rate.
double MinTeacherForcingRatio { get; set; } Gets or sets the minimum teacher forcing ratio.
int DiffusionSteps { get; set; } Gets or sets the number of diffusion steps (K).
int ForecastHorizon { get; set; } Gets or sets the forecast horizon.
int HiddenDimension { get; set; } Gets or sets the hidden dimension for the DCGRU cells.
int MaxDiffusionStep { get; set; } Gets or sets the maximum diffusion step power.
int NumDecoderLayers { get; set; } Gets or sets the number of decoder DCGRU layers.
int NumEncoderLayers { get; set; } Gets or sets the number of encoder DCGRU layers.
int NumNodes { get; set; } Gets or sets the number of nodes in the graph.
int NumSamples { get; set; } Gets or sets the number of samples for uncertainty estimation.
int ScheduledSamplingDecaySteps { get; set; } Gets or sets the number of training steps for scheduled sampling decay.
int SequenceLength { get; set; } Gets or sets the sequence length (input time steps).
string FilterType { get; set; } Gets or sets the type of filter for diffusion.