Causal ML Demos

This repository contains small demos illustrating fundamental concepts behind causal machine learning and statistical learning.

This repository is part of an ongoing exploration of causal machine learning and causal NLP, focusing on how models behave under distribution shift and spurious correlations.

The goal of this repository is to build intuition through minimal runnable experiments that connect:

• probability theory
• statistical inference
• causal reasoning
• distribution shift
• representation learning

Each demo focuses on a single core concept and provides a small reproducible example.

---

Demos

1. Search

1.1 Toy Search (A* planning demo)

search/toy_search A small A* search demo illustrating planning under fixed world assumptions and how such systems may fail under mechanism shift.

---

2. Probability

2.1 Law of Large Numbers (LLN Simulation)

probability/lln_demo A simulation illustrating how empirical averages converge to the true expectation as the number of samples increases.

This principle underlies:

• statistical estimation
• empirical risk minimization
• machine learning training procedures

2.2 Central Limit Theorem (CLT Simulation)

probability/clt_demo A simulation showing how the distribution of the sample mean approaches a normal distribution as the sample size increases.

2.3 Bootstrap Confidence Intervals

bootstrap_ci_demo A simulation illustrating how bootstrap resampling approximates the sampling distribution of an estimator and enables confidence interval estimation.

---

Repository Structure

causal-ml-demos
│
├── search
│   └── toy_search
│
├── probability
│   ├── lln_demo
│   ├── clt_demo
│   └── bootstrap_ci_demo
│
└── README.md

---

Why this repository

Modern machine learning systems often rely on statistical correlations learned from data.

However, robust generalization under distribution shift requires understanding the underlying causal mechanisms that generate data.

These demos aim to explore the conceptual foundations behind:

• causality
• invariance
• mechanism shift
• shortcut learning

through simple and reproducible experiments.