Module causal_inference 

Causal Inference Engine — do-calculus, interventional distributions, and counterfactual reasoning over Gaussian structural equation models.

Overview

This module implements a full causal inference pipeline built on directed acyclic graphs (DAGs) with Gaussian node semantics. The design follows Pearl’s do-calculus framework:

1. Structural Causal Model (SCM) — each node is parameterised by a mean and a variance under a Gaussian structural equation model.
2. Interventional inference (do_calculus) — cuts incoming edges to the intervened node and propagates the fixed value through all directed paths to the target, accumulating linear causal effects.
3. Counterfactual inference (counterfactual) — applies an intervention and conditions on observed evidence by adding a weighted correction from each evidence node to the target.
4. Average Causal Effect (average_causal_effect) — contrasts interventional distributions to quantify treatment effects.
5. d-separation — checks whether two nodes are conditionally independent given a set of observed variables.
6. Backdoor paths — enumerates confounding paths for identifiability analysis.

Structs

CausalEdge

A directed edge in the causal graph carrying a linear strength coefficient.

CausalGraph

The underlying directed acyclic graph over causal nodes.

CausalInferenceEngine

Production-grade causal inference engine.

CausalNode

A variable in the structural causal model parameterised by a Gaussian prior.

CausalNodeId

Opaque identifier for a node inside a CausalGraph.

CausalStats

Summary statistics about the structure of a CausalGraph.

CounterfactualQuery

A counterfactual query: what would target be if we had applied intervention, given that we observed evidence?

InferenceResult

The result returned by an inference query.

Intervention

A hard intervention: set node node to exactly value (written do(X = value)).

Enums

CausalEdgeType

Semantic classification for an edge in a CausalGraph.

CausalError

Errors that can be produced by the CausalInferenceEngine.