Struct StructuralCausalModel 

pub struct StructuralCausalModel {
    pub graph: CausalGraph,
    pub coefficients: Vec<Vec<f64>>,
    pub noise_std: Vec<f64>,
    pub intercepts: Vec<f64>,
}

A linear structural causal model (SCM).

Each variable X_i is defined as: X_i = Σ_j (coeff[i][j] * X_j) + noise_std[i] * ε_i

where ε_i ~ N(0,1) and j ranges over parents of i.

Fields

graph: CausalGraph

The underlying causal graph.

coefficients: Vec<Vec<f64>>

Structural coefficients: coefficients[i][k] is the coefficient for the k-th parent of node i.

noise_std: Vec<f64>

Standard deviation of the exogenous noise for each variable.

intercepts: Vec<f64>

Intercept terms for each variable.

Implementations

impl StructuralCausalModel

pub fn new(n: usize) -> Self

Create a new linear SCM on n variables.

All coefficients are zero, noise std = 1.0, intercepts = 0.0 by default.

pub fn add_edge(&mut self, from: usize, to: usize, coeff: f64)

Add a causal edge with a specified structural coefficient.

Arguments

• from — parent (cause) node index
• to — child (effect) node index
• coeff — linear coefficient

pub fn set_noise(&mut self, v: usize, std: f64)

Set the noise standard deviation for variable v.

pub fn set_intercept(&mut self, v: usize, intercept: f64)

Set the intercept for variable v.

pub fn sample_with_noise(&self, noise: &[f64]) -> Vec<f64>

Sample one observation from the SCM using provided noise values.

Arguments

• noise — exogenous noise values ε_i for each variable (length = n_nodes)

Returns x[i] values in topological order.

pub fn intervene(&self, target: usize, val: f64) -> Self

Perform a do-calculus intervention: set variable target to value val.

Returns the modified SCM where all incoming edges to target are removed and its value is fixed at val (zero noise, intercept = val).

pub fn average_causal_effect( &self, cause: usize, val: f64, effect: usize, noise_samples: &[Vec<f64>], ) -> f64

Compute the average causal effect (ACE) of intervention do(X_cause = val) on variable effect, using the provided noise samples.

Arguments

• cause — the variable to intervene on
• val — the intervention value
• effect — the outcome variable
• noise_samples — matrix of noise samples, shape [n_samples][n_nodes]

pub fn total_effect_linear(&self, cause: usize, effect: usize) -> f64

Compute the total causal effect of cause on effect analytically (only valid for linear SCMs).

Sums all directed path contributions.

Trait Implementations

impl Clone for StructuralCausalModel

fn clone(&self) -> StructuralCausalModel

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for StructuralCausalModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.