CSDIOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for CSDI (Conditional Score-based Diffusion model for Imputation).

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

For Beginners

CSDI solves a critical problem in real-world data: missing values. Instead of simple interpolation, it generates plausible values that are consistent with the observed data.

The Key Insight: Unlike TimeGrad which forecasts future values, CSDI focuses on imputation: filling in missing values WITHIN the observed time series. It conditions on what you DO know to infer what you DON’T know.

How CSDI Works:

  1. Conditional Masking: Identify which values are observed vs missing
  2. Score Matching: Learn the gradient of log probability (the “score”)
  3. Reverse Diffusion: Start from noise, gradually denoise conditioned on observed values
  4. Imputation: Generate multiple samples for uncertainty quantification

Architecture:

  • Transformer-based score network with self-attention
  • Temporal and feature embeddings for position encoding
  • Conditional U-Net style residual blocks
  • Side information integration for covariates

Key Benefits:

  • Handles arbitrary missing patterns (not just regular gaps)
  • Provides uncertainty estimates for imputed values
  • Can incorporate side information (covariates)
  • State-of-the-art imputation quality

How It Works

CSDI is a probabilistic model for time series imputation that uses score-based diffusion to fill in missing values with well-calibrated uncertainty estimates.

Reference: Tashiro et al., “CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation”, 2021. https://arxiv.org/abs/2107.03502

Constructors

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

Properties

Property Summary
bool UseAttention { get; set; } Gets or sets whether to use self-attention in the score network.
bool UseSideInfo { get; set; } Gets or sets whether to use side information (covariates).
double BetaEnd { get; set; } Gets or sets the ending noise level (beta_T).
double BetaStart { get; set; } Gets or sets the starting noise level (beta_1).
double DropoutRate { get; set; } Gets or sets the dropout rate for regularization.
int FeatureEmbeddingDim { get; set; } Gets or sets the dimension of feature embeddings.
int HiddenDimension { get; set; } Gets or sets the hidden dimension for the score network.
int NumDiffusionSteps { get; set; } Gets or sets the number of diffusion steps.
int NumHeads { get; set; } Gets or sets the number of attention heads in the transformer layers.
int NumResidualLayers { get; set; } Gets or sets the number of residual layers in the score network.
int NumSamples { get; set; } Gets or sets the number of samples to generate for uncertainty estimation.
int SequenceLength { get; set; } Gets or sets the sequence length (time steps).
int SideInfoDim { get; set; } Gets or sets the dimension of side information features.
int TimeEmbeddingDim { get; set; } Gets or sets the dimension of time step embeddings.
string BetaSchedule { get; set; } Gets or sets the noise schedule type.