GANITE

Reference: Jinsung Yoon, James Jordon, Mihaela van der Schaar, GANITE: Estimation of Individualized Treatment Effects using Generative Adversarial Nets, ICLR 2018

Implementation remarks: our implementation is exactly the same of the original paper.

GANITE on IHDP

from causalforge.model import Model , PROBLEM_TYPE
from causalforge.data_loader import DataLoader

# load IHDP dataset
r = DataLoader.get_loader('IHDP').load()
X_tr, T_tr, YF_tr, YCF_tr, mu_0_tr, mu_1_tr, X_te, T_te, YF_te, YCF_te, mu_0_te, mu_1_te = r

# model
params={}
params['dim_hidden'] = 100
params['minibatch_size'] = 256
params['alpha'] = 0.1
params['beta'] = 0
params['num_iterations'] = 500
params['num_discr_iterations'] = 1


ganite = Model.create_model("ganite",
                            params,
                            problem_type=PROBLEM_TYPE.CAUSAL_TREATMENT_EFFECT_ESTIMATION,
                            multiple_treatments=False)

Parameters

• dim_hidden: hidden dimension (by default, 100)

• minibatch_size: batch-size (by default, 256)

• reg_l2: L2 regularization (by default, 0.01)

• alpha: alpha (by default, 0.1)

• beta: beta (by default, 0)

• depth: depth (by default, 0)

• num_iterations: number of iterations (by default, 5000)

• num_discr_iterations: number of iterations of discriminator (by default, 1)

Training

from causalforge.metrics import eps_ATE_diff, PEHE_with_ite
import numpy as np

experiment_ids = [1,10,400]

eps_ATE_tr, eps_ATE_te = [], []
eps_PEHE_tr, eps_PEHE_te = [] , []



for idx in experiment_ids:
    t_tr, y_tr, x_tr, mu0tr, mu1tr = T_tr[:,idx] , YF_tr[:,idx], X_tr[:,:,idx], mu_0_tr[:,idx], mu_1_tr[:,idx]
    t_te, y_te, x_te, mu0te, mu1te = T_te[:,idx] , YF_te[:,idx], X_te[:,:,idx], mu_0_te[:,idx], mu_1_te[:,idx]


    # Train your causal method on train-set ...
    ganite.fit(x_tr,t_tr,y_tr)

    # Validate your method test-set ...
    ATE_truth_tr = (mu1tr - mu0tr).mean()
    ATE_truth_te = (mu1te - mu0te).mean()

    ITE_truth_tr = (mu1tr - mu0tr)
    ITE_truth_te = (mu1te - mu0te)

    eps_ATE_tr.append( eps_ATE_diff( ganite.predict_ate(x_tr,t_tr,y_tr), ATE_truth_tr) )
    eps_ATE_te.append( eps_ATE_diff( ganite.predict_ate(x_te,t_te,y_te), ATE_truth_te) )

    eps_PEHE_tr.append( PEHE_with_ite( ganite.predict_ite(x_tr), ITE_truth_tr, sqrt=True))
    eps_PEHE_te.append( PEHE_with_ite( ganite.predict_ite(x_te), ITE_truth_te , sqrt=True))

Generator loss epoch 0: 0.6381762623786926 30.862741470336914
Generator loss epoch 100: 0.003728393232449889 16.405466079711914
Generator loss epoch 200: 0.0014365405077114701 9.917235374450684
Generator loss epoch 300: 0.0011493910569697618 6.902956008911133
Generator loss epoch 400: 0.0013534935424104333 3.959357738494873
Inference loss epoch 0: 21.016639709472656
Inference loss epoch 100: 15.584854125976562
Inference loss epoch 200: 9.773422241210938
Inference loss epoch 300: 8.935896873474121
Inference loss epoch 400: 7.52246618270874
Generator loss epoch 0: 0.6245900392532349 40.30231857299805
Generator loss epoch 100: 0.005318560171872377 22.773630142211914
Generator loss epoch 200: 0.0017221181187778711 14.723432540893555
Generator loss epoch 300: 0.001508277957327664 8.983695983886719
Generator loss epoch 400: 0.0011473450576886535 5.9448418617248535
Inference loss epoch 0: 23.86356544494629
Inference loss epoch 100: 15.311829566955566
Inference loss epoch 200: 13.44272518157959
Inference loss epoch 300: 9.781745910644531
Inference loss epoch 400: 7.9263916015625
Generator loss epoch 0: 0.5558345913887024 159.0293426513672
Generator loss epoch 100: 0.0016039644833654165 144.30055236816406
Generator loss epoch 200: 0.0006120165344327688 93.21966552734375
Generator loss epoch 300: 0.0006608975236304104 75.96114349365234
Generator loss epoch 400: 0.0006516539724543691 39.48774719238281
Inference loss epoch 0: 79.98333740234375
Inference loss epoch 100: 68.76531982421875
Inference loss epoch 200: 56.01595687866211
Inference loss epoch 300: 56.58964157104492
Inference loss epoch 400: 46.454689025878906

Results

import pandas as pd

pd.DataFrame([[np.mean(eps_ATE_tr),np.mean(eps_ATE_te),np.mean(eps_PEHE_tr),np.mean(eps_PEHE_te)]],
             columns=['eps_ATE_tr','eps_ATE_te','eps_PEHE_tr','eps_PEHE_te'],
             index=['DragonNet'])

	eps_ATE_tr	eps_ATE_te	eps_PEHE_tr	eps_PEHE_te
GANITE	1.289279	1.360971	3.802482	3.843936

ITE distribution: learned vs. ground truth

Learned

from causalforge.utils import plot_ite_distribution

plot_ite_distribution(ganite.predict_ite(x_te))

Ground Truth

from causalforge.utils import plot_ite_distribution

plot_ite_distribution(ITE_truth_te)