BCAUSS
BCAUSS 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['input_dim'] = X_tr.shape[1]
bcauss = Model.create_model("bcauss",
params,
problem_type=PROBLEM_TYPE.CAUSAL_TREATMENT_EFFECT_ESTIMATION,
multiple_treatments=False)
bcauss.model.summary()
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input (InputLayer) [(None, 25)] 0 []
dense (Dense) (None, 200) 5200 ['input[0][0]']
dense_1 (Dense) (None, 200) 40200 ['dense[0][0]']
dense_2 (Dense) (None, 200) 40200 ['dense_1[0][0]']
dense_4 (Dense) (None, 100) 20100 ['dense_2[0][0]']
dense_5 (Dense) (None, 100) 20100 ['dense_2[0][0]']
dense_6 (Dense) (None, 100) 10100 ['dense_4[0][0]']
dense_7 (Dense) (None, 100) 10100 ['dense_5[0][0]']
dense_3 (Dense) (None, 1) 201 ['dense_2[0][0]']
y0_predictions (Dense) (None, 1) 101 ['dense_6[0][0]']
y1_predictions (Dense) (None, 1) 101 ['dense_7[0][0]']
epsilon_layer (EpsilonLayer) (None, 1) 1 ['dense_3[0][0]']
y_true (InputLayer) [(None, 1)] 0 []
t_true (InputLayer) [(None, 1)] 0 []
concatenate (Concatenate) (None, 4) 0 ['y0_predictions[0][0]',
'y1_predictions[0][0]',
'dense_3[0][0]',
'epsilon_layer[0][0]']
tf.__operators__.add (TFOpLamb (None, 1) 0 ['dense_3[0][0]']
da)
tf.math.truediv (TFOpLambda) (None, 1) 0 ['tf.__operators__.add[0][0]']
tf.math.subtract_7 (TFOpLambda (None, 1) 0 ['t_true[0][0]']
)
tf.math.subtract_8 (TFOpLambda (None, 1) 0 ['tf.math.truediv[0][0]']
)
tf.math.truediv_3 (TFOpLambda) (None, 1) 0 ['t_true[0][0]',
'tf.math.truediv[0][0]']
tf.math.truediv_4 (TFOpLambda) (None, 1) 0 ['tf.math.subtract_7[0][0]',
'tf.math.subtract_8[0][0]']
tf.split (TFOpLambda) [(None, 1)] 0 ['tf.math.truediv_3[0][0]']
tf.split_1 (TFOpLambda) [(None, 1)] 0 ['tf.math.truediv_4[0][0]']
tf.concat (TFOpLambda) (None, 25) 0 ['tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]',
'tf.split[0][0]']
tf.concat_1 (TFOpLambda) (None, 25) 0 ['tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]',
'tf.split_1[0][0]']
tf.math.subtract_9 (TFOpLambda (None, 1) 0 ['t_true[0][0]']
)
tf.math.subtract_10 (TFOpLambd (None, 1) 0 ['tf.math.truediv[0][0]']
a)
tf.math.multiply_4 (TFOpLambda (None, 25) 0 ['tf.concat[0][0]',
) 'input[0][0]']
tf.math.truediv_5 (TFOpLambda) (None, 1) 0 ['t_true[0][0]',
'tf.math.truediv[0][0]']
tf.math.multiply_5 (TFOpLambda (None, 25) 0 ['tf.concat_1[0][0]',
) 'input[0][0]']
tf.math.truediv_7 (TFOpLambda) (None, 1) 0 ['tf.math.subtract_9[0][0]',
'tf.math.subtract_10[0][0]']
tf.math.reduce_sum_3 (TFOpLamb (25,) 0 ['tf.math.multiply_4[0][0]']
da)
tf.math.reduce_sum_4 (TFOpLamb (1,) 0 ['tf.math.truediv_5[0][0]']
da)
tf.math.reduce_sum_5 (TFOpLamb (25,) 0 ['tf.math.multiply_5[0][0]']
da)
tf.math.reduce_sum_6 (TFOpLamb (1,) 0 ['tf.math.truediv_7[0][0]']
da)
tf.math.subtract_1 (TFOpLambda (None, 1) 0 ['y_true[0][0]',
) 'y0_predictions[0][0]']
tf.math.subtract_2 (TFOpLambda (None, 1) 0 ['y_true[0][0]',
) 'y1_predictions[0][0]']
tf.math.truediv_6 (TFOpLambda) (25,) 0 ['tf.math.reduce_sum_3[0][0]',
'tf.math.reduce_sum_4[0][0]']
tf.math.truediv_8 (TFOpLambda) (25,) 0 ['tf.math.reduce_sum_5[0][0]',
'tf.math.reduce_sum_6[0][0]']
tf.math.subtract (TFOpLambda) (None, 1) 0 ['t_true[0][0]']
tf.math.square (TFOpLambda) (None, 1) 0 ['tf.math.subtract_1[0][0]']
tf.math.square_1 (TFOpLambda) (None, 1) 0 ['tf.math.subtract_2[0][0]']
tf.convert_to_tensor_2 (TFOpLa (25,) 0 ['tf.math.truediv_6[0][0]']
mbda)
tf.cast (TFOpLambda) (25,) 0 ['tf.math.truediv_8[0][0]']
tf.math.multiply (TFOpLambda) (None, 1) 0 ['tf.math.subtract[0][0]',
'tf.math.square[0][0]']
tf.math.multiply_1 (TFOpLambda (None, 1) 0 ['t_true[0][0]',
) 'tf.math.square_1[0][0]']
tf.math.squared_difference (TF (25,) 0 ['tf.convert_to_tensor_2[0][0]',
OpLambda) 'tf.cast[0][0]']
tf.math.reduce_sum_1 (TFOpLamb () 0 ['tf.math.multiply[0][0]']
da)
tf.math.reduce_sum_2 (TFOpLamb () 0 ['tf.math.multiply_1[0][0]']
da)
tf.math.reduce_mean (TFOpLambd () 0 ['tf.math.squared_difference[0][0
a) ]']
tf.__operators__.add_1 (TFOpLa () 0 ['tf.math.reduce_sum_1[0][0]',
mbda) 'tf.math.reduce_sum_2[0][0]']
tf.math.multiply_6 (TFOpLambda () 0 ['tf.math.reduce_mean[0][0]']
)
tf.__operators__.add_3 (TFOpLa () 0 ['tf.__operators__.add_1[0][0]',
mbda) 'tf.math.multiply_6[0][0]']
add_loss (AddLoss) () 0 ['tf.__operators__.add_3[0][0]']
==================================================================================================
Total params: 146,404
Trainable params: 146,404
Non-trainable params: 0
__________________________________________________________________________________________________
Parameters
input_dim: number of inputs
neurons_per_layer: number of neurons per layer (by default,
200)act_fn: activation function [
relu|elu|tanh] (by default,relu)reg_l2: L2 regularization (by default,
0.01)verbose: verbose (by default,
True)val_split: validation split ratio (by default,
0.22)ratio: relative importance of the targeted regularization term, if adopted (by default,
1.0)optim: optimizer [
sgd|adam] (by default,sgd)epochs: number of epochs (by default,
500)learning_rate: learning rate (by default,
1e-5)momentum: momentum (by default,
0.9)use_bce: to use the binary cross-entropy to train the propensity score estimator (by default,
False)norm_bal_term: to normalize the auto-balancing regularization term (by default,
True)use_targ_term: to use the targeted regularization objective (by default,
False)b_ratio: relative importance of the auto-balancing regularization objective (by default,
1.0)bs_ratio: percentage of train-set to use as batch-size (by default,
1.0)scale_preds: to scale predictions (by default,
True)
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 ...
bcauss.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( bcauss.predict_ate(x_tr,t_tr,y_tr), ATE_truth_tr) )
eps_ATE_te.append( eps_ATE_diff( bcauss.predict_ate(x_te,t_te,y_te), ATE_truth_te) )
eps_PEHE_tr.append( PEHE_with_ite( bcauss.predict_ite(x_tr), ITE_truth_tr, sqrt=True))
eps_PEHE_te.append( PEHE_with_ite( bcauss.predict_ite(x_te), ITE_truth_te , sqrt=True))
Epoch 1/500 ... Epoch 499/500 1/1 [==============================] - 0s 43ms/step - loss: 14.0257 - val_loss: 16.1863 - lr: 1.0000e-05 Epoch 500/500 1/1 [==============================] - 0s 41ms/step - loss: 14.0177 - val_loss: 16.1839 - lr: 1.0000e-05 ************************* elapsed_time is: 21.736793041229248 21/21 [==============================] - 0s 1ms/step 3/3 [==============================] - 0s 2ms/step 21/21 [==============================] - 0s 1ms/step 3/3 [==============================] - 0s 1ms/step
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=['BCAUSS'])
| eps_ATE_tr | eps_ATE_te | eps_PEHE_tr | eps_PEHE_te | |
|---|---|---|---|---|
| BCAUSS | 0.067121 | 0.084479 | 0.567214 | 0.510539 |
ITE distribution: learned vs. ground truth
Learned
from causalforge.utils import plot_ite_distribution
plot_ite_distribution(bcauss.predict_ite(x_te))
Ground Truth
from causalforge.utils import plot_ite_distribution
plot_ite_distribution(ITE_truth_te)
