uplift_tutorial.py

import numpy as np
from sklearn.linear_model import LinearRegression
from sklearn.ensemble import RandomForestRegressor

num_obs = 1000

t = np.random.binomial(1, .5, num_obs)
x = np.random.binomial(1, .5, num_obs)
y = 10 + 2*t*x - t + np.random.normal(0, .1, num_obs)

#setting up data matrix with interaction
x_data = np.concatenate([q.reshape(-1,1) for q in [x, t, x*t]], axis = 1)

#inspect data 
print(x_data[:5])

#fit regression
reg = LinearRegression()
reg.fit(x_data, y)

#look at coefficients
print(np.round(reg.coef_, 2))
print(np.round(reg.intercept_, 2))



###.fit() -> .predict() Method
x_data_no_interaction = np.concatenate([q.reshape(-1,1) for q in [x, t]], 
  axis = 1)

#inspect data 
print(x_data_no_interaction[:5])


rf = RandomForestRegressor()
rf.fit(x_data_no_interaction, y.reshape(-1,1))

x_data_no_interaction_0 = x_data_no_interaction.copy()
x_data_no_interaction_0[:,1] = 0

x_data_no_interaction_1 = x_data_no_interaction.copy()
x_data_no_interaction_1[:,1] = 1

pred_y_0 = rf.predict(x_data_no_interaction_0)
pred_y_1 = rf.predict(x_data_no_interaction_1)

ite = pred_y_1 - pred_y_0