Skip to content

CausalPipe Documentation

Welcome to the documentation for CausalPipe, a Python package that simplifies causal discovery and analysis. CausalPipe wraps around Causal-Learn and Lavaan to provide an easy-to-use, step-by-step process for performing causal analysis.

Features

  • Data Preprocessing: Handles missing data with multiple imputation, standardizes variables, and encodes categorical variables.
  • Causal Discovery: Construct causal graphs using methods like Fast Adjacency Search (FAS) or Bootstrap-based Causal Structure Learning (BCSL).
  • Edge Orientation: Use algorithms such as Fast Causal Inference (FCI) or Hill Climbing to orient edges in causal graphs.
  • Causal Effect Estimation: Estimate effects using methods such as Partial Correlation, Structural Equation Modeling (SEM), or kernel methods.
  • Symbolic Regression: Integrate PySR to learn nonlinear structural equations and score cyclic models via pseudo-likelihood or MMD2.
  • Visualization: Visualize correlation graphs, causal graphs, and SEM results.

Configuration Classes

CausalPipe is configured through a collection of dataclasses that define each step of the pipeline:

  • VariableTypes – declare continuous, ordinal, and nominal variables.
  • DataPreprocessingParams – control imputation, standardization, and feature filtering.
  • Skeleton methods – FASSkeletonMethod, BCSLSkeletonMethod.
  • Orientation methods – FCIOrientationMethod, HillClimbingOrientationMethod.
  • Causal effect methods – PearsonCausalEffectMethod, SpearmanCausalEffectMethod, MICausalEffectMethod, KCICausalEffectMethod, SEMCausalEffectMethod, SEMClimbingCausalEffectMethod, PYSRCausalEffectMethod, PYSRCausalEffectMethodHillClimbing.