forked from ioanabica/SCIGAN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_simulation.py
199 lines (144 loc) · 8.01 KB
/
data_simulation.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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
# Copyright (c) 2020, Ioana Bica
from __future__ import print_function
import numpy as np
import pickle
from sklearn.model_selection import StratifiedShuffleSplit
def softmax(x):
e_x = np.exp(x)
return e_x / e_x.sum(axis=0)
def compute_beta(alpha, optimal_dosage):
if (optimal_dosage <= 0.001 or optimal_dosage >= 1.0):
beta = 1.0
else:
beta = (alpha - 1.0) / float(optimal_dosage) + (2.0 - alpha)
return beta
def generate_patient(x, v, num_treatments, treatment_selection_bias, dosage_selection_bias,
scaling_parameter, noise_std):
outcomes = []
dosages = []
for treatment in range(num_treatments):
if (treatment == 0):
b = 0.75 * np.dot(x, v[treatment][1]) / (np.dot(x, v[treatment][2]))
if (b >= 0.75):
optimal_dosage = b / 3.0
else:
optimal_dosage = 1.0
alpha = dosage_selection_bias
dosage = np.random.beta(alpha, compute_beta(alpha, optimal_dosage))
y = get_patient_outcome(x, v, treatment, dosage, scaling_parameter)
elif (treatment == 1):
optimal_dosage = np.dot(x, v[treatment][2]) / (2.0 * np.dot(x, v[treatment][1]))
alpha = dosage_selection_bias
dosage = np.random.beta(alpha, compute_beta(alpha, optimal_dosage))
if (optimal_dosage <= 0.001):
dosage = 1 - dosage
y = get_patient_outcome(x, v, treatment, dosage, scaling_parameter)
elif (treatment == 2):
optimal_dosage = np.dot(x, v[treatment][1]) / (2.0 * np.dot(x, v[treatment][2]))
alpha = dosage_selection_bias
dosage = np.random.beta(alpha, compute_beta(alpha, optimal_dosage))
if (optimal_dosage <= 0.001):
dosage = 1 - dosage
y = get_patient_outcome(x, v, treatment, dosage, scaling_parameter)
outcomes.append(y)
dosages.append(dosage)
treatment_coeff = [treatment_selection_bias * (outcomes[i] / np.max(outcomes)) for i in range(num_treatments)]
treatment = np.random.choice(num_treatments, p=softmax(treatment_coeff))
return treatment, dosages[treatment], outcomes[treatment] + np.random.normal(0, noise_std)
def get_patient_outcome(x, v, treatment, dosage, scaling_parameter=10):
if (treatment == 0):
y = float(scaling_parameter) * (np.dot(x, v[treatment][0]) + 12.0 * dosage * (dosage - 0.75 * (
np.dot(x, v[treatment][1]) / np.dot(x, v[treatment][2]))) ** 2)
elif (treatment == 1):
y = float(scaling_parameter) * (np.dot(x, v[treatment][0]) + np.sin(
np.pi * (np.dot(x, v[treatment][1]) / np.dot(x, v[treatment][2])) * dosage))
elif (treatment == 2):
y = float(scaling_parameter) * (np.dot(x, v[treatment][0]) + 12.0 * (np.dot(x, v[treatment][
1]) * dosage - np.dot(x, v[treatment][2]) * dosage ** 2))
return y
def get_dataset_splits(dataset):
dataset_keys = ['x', 't', 'd', 'y', 'y_normalized']
train_index = dataset['metadata']['train_index']
val_index = dataset['metadata']['val_index']
test_index = dataset['metadata']['test_index']
dataset_train = dict()
dataset_val = dict()
dataset_test = dict()
for key in dataset_keys:
dataset_train[key] = dataset[key][train_index]
dataset_val[key] = dataset[key][val_index]
dataset_test[key] = dataset[key][test_index]
dataset_train['metadata'] = dataset['metadata']
dataset_val['metadata'] = dataset['metadata']
dataset_test['metadata'] = dataset['metadata']
return dataset_train, dataset_val, dataset_test
def get_split_indices(num_patients, patients, treatments, validation_fraction, test_fraction):
num_validation_patients = int(np.floor(num_patients * validation_fraction))
num_test_patients = int(np.floor(num_patients * test_fraction))
test_sss = StratifiedShuffleSplit(n_splits=1, test_size=num_test_patients, random_state=0)
rest_indices, test_indices = next(test_sss.split(patients, treatments))
val_sss = StratifiedShuffleSplit(n_splits=1, test_size=num_validation_patients, random_state=0)
train_indices, val_indices = next(val_sss.split(patients[rest_indices], treatments[rest_indices]))
return train_indices, val_indices, test_indices
class TCGA_Data():
def __init__(self, args):
np.random.seed(3)
self.num_treatments = args['num_treatments']
self.treatment_selection_bias = args['treatment_selection_bias']
self.dosage_selection_bias = args['dosage_selection_bias']
self.validation_fraction = args['validation_fraction']
self.test_fraction = args['test_fraction']
self.tcga_data = pickle.load(open('datasets/tcga.p', 'rb'))
self.patients = self.normalize_data(self.tcga_data['rnaseq'])
self.scaling_parameteter = 10
self.noise_std = 0.2
self.num_weights = 3
self.v = np.zeros(shape=(self.num_treatments, self.num_weights, self.patients.shape[1]))
for i in range(self.num_treatments):
for j in range(self.num_weights):
self.v[i][j] = np.random.uniform(0, 10, size=(self.patients.shape[1]))
self.v[i][j] = self.v[i][j] / np.linalg.norm(self.v[i][j])
self.dataset = self.generate_dataset(self.patients, self.num_treatments)
def normalize_data(self, patient_features):
x = (patient_features - np.min(patient_features, axis=0)) / (
np.max(patient_features, axis=0) - np.min(patient_features, axis=0))
for i in range(x.shape[0]):
x[i] = x[i] / np.linalg.norm(x[i])
return x
def generate_dataset(self, patient_features, num_treatments):
tcga_dataset = dict()
tcga_dataset['x'] = []
tcga_dataset['y'] = []
tcga_dataset['t'] = []
tcga_dataset['d'] = []
tcga_dataset['metadata'] = dict()
tcga_dataset['metadata']['v'] = self.v
tcga_dataset['metadata']['treatment_selection_bias'] = self.treatment_selection_bias
tcga_dataset['metadata']['dosage_selection_bias'] = self.dosage_selection_bias
tcga_dataset['metadata']['noise_std'] = self.noise_std
tcga_dataset['metadata']['scaling_parameter'] = self.scaling_parameteter
for patient in patient_features:
t, dosage, y = generate_patient(x=patient, v=self.v, num_treatments=num_treatments,
treatment_selection_bias=self.treatment_selection_bias,
dosage_selection_bias=self.dosage_selection_bias,
scaling_parameter=self.scaling_parameteter,
noise_std=self.noise_std)
tcga_dataset['x'].append(patient)
tcga_dataset['t'].append(t)
tcga_dataset['d'].append(dosage)
tcga_dataset['y'].append(y)
for key in ['x', 't', 'd', 'y']:
tcga_dataset[key] = np.array(tcga_dataset[key])
tcga_dataset['metadata']['y_min'] = np.min(tcga_dataset['y'])
tcga_dataset['metadata']['y_max'] = np.max(tcga_dataset['y'])
tcga_dataset['y_normalized'] = (tcga_dataset['y'] - np.min(tcga_dataset['y'])) / (
np.max(tcga_dataset['y']) - np.min(tcga_dataset['y']))
train_indices, validation_indices, test_indices = get_split_indices(num_patients=tcga_dataset['x'].shape[0],
patients=tcga_dataset['x'],
treatments=tcga_dataset['t'],
validation_fraction=self.validation_fraction,
test_fraction=self.test_fraction)
tcga_dataset['metadata']['train_index'] = train_indices
tcga_dataset['metadata']['val_index'] = validation_indices
tcga_dataset['metadata']['test_index'] = test_indices
return tcga_dataset