This is the code repository for the paper Causal de Finetti: On the identification of invariant causal structure in Exchangeable data (NeurIPS 2023).
Performing Causal-de-Finetti algorithm on exchagneable data allows one to uniquely identify the correct causal structure in bivariate and multivariate settings.
To reproduce the results, pip install the requirements.txt file and
run the main.py file. The results are saved in experiments/results folder.
To run the causal de Finetti algorithm is as easy as running the following code snippet:
from src.models.causaldf import *
from experiments.synthetic_data_generation import *
num_env_bivariate = 1000
num_env_multivariate = 5000
num_sample_per_env = 2
## Bivariate
data = scm_bivariate_continuous(num_env = num_env_bivariate, num_sample = num_sample_per_env)
estimate, _ = run_causaldf_bivariate(data)
correct = True if estimate == data['true_structure'] else False
print('Is correct in bivariate:', correct)
## Multivariate
data = scm_multivariate_binary(num_env = num_env_multivariate, num_sample = num_sample_per_env, num_var = 3)
estimate, _ = run_causaldf_multivariate(data)
correct = True if estimate == data['true_structure'] else False
print('Is correct in multivariate:', correct)
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent