-
Notifications
You must be signed in to change notification settings - Fork 3
/
Simulation_analyse.py
67 lines (52 loc) · 3.32 KB
/
Simulation_analyse.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
import numpy as np
from bites.analyse.analyse_utils import *
import pickle
import matplotlib.pyplot as plt
if __name__ == '__main__':
"""Choose Method to analyse!"""
###############################
method = 'DeepSurv' # Set Name: BITES, ITES, DeepSurv, DeepSurvT, CFRNet
results_dir='ray_results/' # Set result_dir
num_training_samples = 0 # int 0 to 4 for 1000 to 4000 training samples
trial_name = 'Simulation3' # Name used in the config
################################
"""Load triaining (for baseline hazards) and test data."""
X_test, y_test = pickle.load(open('data/Simulation_Treatment_Bias/test_data.Sim3', 'rb'))
X_train, y_train = pickle.load(open('data/Simulation_Treatment_Bias/train_data.Sim3', 'rb'))[0]
Y_test, event_test, treatment_test = y_test[:, 5], y_test[:, 4], y_test[:, 2]
# Analysis of the different Methods
result_path = results_dir + method + "_" + trial_name
pred_ite=None
if method == 'BITES' or method == 'ITES':
model, config = get_best_model(result_path)
model.compute_baseline_hazards(X_train, [y_train[:, 5], y_train[:, 4], y_train[:, 2]])
C_index, C_index_T0, C_index_T1 = get_C_Index_BITES(model, X_test, Y_test, event_test, treatment_test)
pred_ite, correct_predicted_probability = get_ITE_BITES(model, X_test, treatment_test, best_treatment = y_test[:,3])
elif method == 'DeepSurvT':
model0, config0 = get_best_model(results_dir + method + "_T0_" + trial_name, assign_treatment=0)
model0.compute_baseline_hazards(X_train, [y_train[:, 5], y_train[:, 4], y_train[:, 2]])
model1, config1 = get_best_model(results_dir + method + "_T1_" + trial_name, assign_treatment=1)
model1.compute_baseline_hazards(X_train, [y_train[:, 5], y_train[:, 4], y_train[:, 2]])
C_index, C_index_T0, C_index_T1 = get_C_Index_DeepSurvT(model0, model1, X_test, Y_test, event_test, treatment_test)
pred_ite, correct_predicted_probability = get_ITE_DeepSurvT(model0, model1, X_test, treatment_test, best_treatment=y_test[:,3],
death_probability=0.5)
elif method == 'DeepSurv':
treatment_train = y_train[:, 2]
if treatment_train is not None:
X_train=np.c_[X_train, treatment_train]
X_test=np.c_[X_test, treatment_test]
model, config = get_best_model(result_path)
model.compute_baseline_hazards(X_train, [y_train[:, 5], y_train[:, 4], y_train[:, 2]])
C_index, C_index_T0, C_index_T1 = get_C_Index_DeepSurv(model, X_test, Y_test, event_test)
pred_ite, correct_predicted_probability = get_ITE_DeepSurv(model, X_test, treatment_test, best_treatment=y_test[:,3], death_probability=0.5)
else:
print("No treatment set, return best possible model")
model, config = get_best_model(result_path)
model.compute_baseline_hazards(X_train, [y_train[:, 5], y_train[:, 4], None])
elif method == 'CFRNet':
model, config = get_best_model(result_path)
pred_ite = get_ITE_CFRNet(model, X_test, treatment_test, best_treatment=None)
else:
print(method+' Not defined!')
if pred_ite is not None:
plot_ITE_correlation(pred_ite, y_true=y_test[:,0],y_cf=y_test[:,1],treatment=treatment_test)