Causal Ladder

Usage Guide

1. Create a virtual environment

python -m venv env

2. Activate the virtual environment

For macOS / Linux

source env/bin/activate

For Windows

env\Scripts\Activate.ps1

3. Install pip-tools

pip install pip-tools

4. Install project dependencies

pip-sync

---

A. Generate Synthetic Data

i. Vision Dataset

python -m scripts.generate_synthetic_vision_data --n_samples 10000 --output_filename vision_data.csv --data_dir data

ii. Kidney Stone Dataset

python -m scripts.generate_synthetic_kidney_stone_data --output_filename kidney_data.csv --data_dir data

---

B. Run Rung 1 (Association)

python -m scripts.run_rung1 [FLAGS]

Flags

• --data_path: Path to CSV data file. Default: ./data/kidney_data.csv
• --treatment_col: Column name for treatment. Default: Treatment
• --condition_col: Column name for condition. Default: Stone_Size
• --outcome_col: Column name for outcome. Default: Success
• --stratify: If provided, results are stratified by the condition variable.

Example

python -m scripts.run_rung1 --data_path ./data/kidney_data.csv --stratify

---

C. Run Rung 2 (Intervention)

python -m scripts.run_rung2 [FLAGS]

Flags

• --data_path: Path to CSV data file. Default: ./data/kidney_data.csv
• --random_seed: Seed for reproducibility. Default: 42
• --treatment_col: Column name for treatment. Default: Treatment
• --condition_col: Column name for condition. Default: Stone_Size
• --outcome_col: Column name for outcome. Default: Success

Example

python -m scripts.run_rung2 --data_path ./data/kidney_data.csv --treatment_col Treatment --outcome_col Success

---

D. Run Rung 3 (Counterfactual)

python -m scripts.run_rung3 [FLAGS]

Flags

• --data_path: Path to CSV data file. Default: ./data/vision_data.csv
• --sample_condition: The specific condition value of the patient. Default: 1
• --sample_treatment: The actual treatment the patient received. Default: 2
• --sample_outcome: The actual outcome the patient experienced. Default: 0.883874
• --new_treatments: Space-separated list of alternative treatments to simulate. Default: 0 1
• --treatment_col: Column name for treatment. Default: Treatment
• --condition_col: Column name for condition. Default: Condition
• --outcome_col: Column name for outcome. Default: Vision

Example

python -m scripts.run_rung3 --data_path ./data/vision_data.csv --sample_condition 1 --sample_treatment 2 --sample_outcome 0.88 --new_treatments 0 1

---

References

• Vision Model Logic: Inspired by the DoWhy GCM Counterfactual Medical Example.
• Kidney Stone Dataset: Based on the classical study: Comparison of treatment of renal calculi by open surgery, percutaneous nephrolithotomy, and extracorporeal shockwave lithotripsy (1986).