Causal Analysis Toolkit

Overview

The Causal Analysis Toolkit provides a comprehensive pipeline for:

• Dataset Preprocessing
• Exploratory Data Analysis (EDA) : Histograms and scatterplots
• Pairwise Dependence Analysis : Using Kernel-based Conditional Independence (KCI), Randomized Conditional Independence Test (RCIT), and HSIC tests
• Causal Discovery : Using the PC algorithm from causal-learn, as well as GES.

This repository aims to provide researchers with an easy-to-use framework for causal discovery from observational data.

---

Features

✅ Preprocessing : Handles dataset loading and cleaning

✅ Visualizations : Generates histograms and pairwise scatterplots

✅ Statistical Independence Tests : Computes CI p-values for multiple tests

✅ Causal Discovery : Uses the PC/GES algorithm to infer causal graphs

---

Installation

1️⃣ Clone the Repository

git clone https://github.com/lokali/causal_analysis_toolkit.git
cd causal_analysis_toolkit

2️⃣ Install Dependencies

pip install -r requirements.txt

---

Usage

1️⃣ Run Exploratory Data Analysis

jupyter notebook 01_data_analysis.ipynb

This notebook will:

• Load and visualize the dataset
• Generate histograms and scatterplots
• Compute KCI, RCIT, and HSIC dependence matrices

2️⃣ Perform Causal Discovery

Run run_pc in 01_data_analysis.ipynb and or directly run the pc algorithm from causal-learn:

from utils import run_pc
cg, path = run_pc(data=df.values, alpha=0.01, indep_test='fisherz', label=df.columns.values)

or 

from causallearn.search.ConstraintBased.PC import pc
cg = pc(df.values, alpha=0.01, indep_test="fisherz")

This will infer the causal structure and plot the causal graph.

---

File Structure

causal_analysis_toolkit/
│── 01_data_analysis.ipynb   # Jupyter Notebook for data analysis
│── utils.py                 # Utility functions for CI tests & visualization
│── requirements.txt          # List of dependencies
│── README.md                # Documentation
│── results/                 # Output directory for figures & logs

---

Contact

For questions or feedback, feel free to contact me via Longkang.Li@mbzuai.ac.ae.