PCMCIAlgorithm<T>
Models & Types · AiDotNet.CausalDiscovery.TimeSeries
PCMCI — PC algorithm for Momentary Conditional Independence in time series.
Inheritance: CausalDiscoveryBase<T> → TimeSeriesCausalBase<T> → PCMCIAlgorithm<T>
For Beginners
PCMCI is designed for time series where you want to know which variables’ past values help predict other variables’ current values. It’s more reliable than simple Granger causality because it conditions on the right set of variables.
How It Works
PCMCI combines a condition-selection step (based on PC’s skeleton discovery) with momentary conditional independence (MCI) tests. It first identifies the relevant lagged parents of each variable, then tests for direct causal links conditioned on those parents.
Algorithm:
- PC₁: For each variable, identify candidate lagged parents using iterative CI tests
- MCI: Test X(t-τ) → Y(t) conditioned on parents of both X and Y
Reference: Runge et al. (2019), “Detecting and Quantifying Causal Associations in Large Nonlinear Time Series Datasets”, Science Advances.
Constructors
| Constructor | Summary |
|---|---|
PCMCIAlgorithm(CausalDiscoveryOptions options) |
Properties
| Property | Summary |
|---|---|
bool SupportsNonlinear { get; } |
|
string Name { get; } |