LiNGAM

lingam python code(lingam.py) You can find causal structure of variables using ICA-LiNGAM.

Reference: LiNGAM paper, Qiita(Japanese)

1. Ready data that you want to know causal structure.(LiNGAM only can use Continuous variables.)

import numpy as np
import pandas as pd

size = 10000
np.random.seed(2017)
x = np.random.uniform(size=size)

np.random.seed(1028)
y = 3*x + np.random.uniform(size=size)

X = pd.DataFrame(np.asarray([x,y]).T,columns=["x","y"])

2. Use LiNGAM. Select ICA method kurtosis(default) or negentropy(sklearn).

use kurtosis-based ICA

lingam = LiNGAM()
lingam.fit(X)

or

use negentropy-based ICA

lingam = LiNGAM()
lingam.fit(X,use_sklearn=True)

And, get result(same result).

x ---|2.991|---> y

array([[ 0.        ,  0.        ],
       [ 2.99149033,  0.        ]])

This means correct result.

   x =    0*x + 0*y + e1 = e1

   y = 2.99*x + 0*y + e2 = 2.99x + e2