-
Notifications
You must be signed in to change notification settings - Fork 0
/
arm_distribution_generator.py
159 lines (129 loc) · 5.58 KB
/
arm_distribution_generator.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
import numpy as np
import matplotlib.pyplot as plt
plt.ioff()
import matplotlib.patches as mpatches
from util import pickle_read
from sklearn.preprocessing import normalize
from sklearn.preprocessing import StandardScaler
# find pareto arms from reward vectors
def pareto(matrix, K):
non_dominated= np.ones([K])
dominated= np.zeros([K])
for i in range(K):
for j in range(i+1, K):
if np.all(matrix[i] <= matrix[j]):
non_dominated[i] = 0
dominated[i] = 1
elif np.all(matrix[j] <= matrix[i]):
non_dominated[j] = 0
dominated[j]= 1
return non_dominated.nonzero()[0], dominated.nonzero()[0]
class ArmGenerator:
def __init__(self, K, M, dist_name):
self.K= K
self.dist_name= dist_name
self.M= M
def create_medians_2obj(self, reward_min, reward_max):
K, M = self.K, self.M
median_matrix= np.random.uniform(reward_min,reward_max,(K, M))
self.median_matrix = median_matrix
def create_samples(self, arm_ind, obj_ind, N, epsilon, pareto_inds,std):
indicator = np.random.binomial(1, epsilon, size= N)
median_matrix= self.median_matrix
# contam_amp = 1000
contam_amp = 1
if arm_ind in pareto_inds:
contamination= -contam_amp
else:
contamination = contam_amp
if self.dist_name == 'Gaussian':
samples= np.random.normal(median_matrix[arm_ind][obj_ind], size= N, scale= std) \
* ( 1-indicator) + indicator * contamination
return samples
def create_samples_diabetes(self, arm_ind, obj_ind, N, epsilon, sigma, pareto_inds):
indicator = np.random.binomial(1, epsilon, size= N)
median_matrix= self.y
contam_amp = np.random.uniform(-50, 50)
if arm_ind in pareto_inds:
contamination= -contam_amp
else:
contamination = contam_amp
if self.dist_name == 'Gaussian':
samples= np.random.normal(median_matrix[arm_ind][obj_ind], size= N, scale= sigma) \
+ indicator * contamination
return samples
def create_samples_llvm(self, arm_ind, obj_ind, N, epsilon, pareto_inds):
# contam_amp = 1000
# contam_amp = 1
contam_amp = 10
if arm_ind in pareto_inds:
contamination= -contam_amp
else:
contamination = contam_amp
indicator = np.random.binomial(1, epsilon, size= N)
sample_inds= self.sample_inds_dict[arm_ind]
random_sample_inds= sample_inds[np.random.choice(len(sample_inds), size= N, replace= True)]
true_samples =self.y[random_sample_inds][:, obj_ind]
corrupted_samples = true_samples * (1- indicator) + indicator* contamination
return corrupted_samples
def load_llvm(self, file_name):
llvm_dict= pickle_read(file_name)
self.y = llvm_dict['y']
self.x= llvm_dict['x']
self.sample_inds_dict= llvm_dict['sample_inds_dict']
median_matrix= np.zeros([0,2])
for arm in self.sample_inds_dict:
mean_arm = np.mean(self.y[self.sample_inds_dict[arm], :], axis= 0, keepdims= True)
median_matrix = np.append(median_matrix, mean_arm, axis = 0)
self.median_matrix = median_matrix
std_matrix = np.zeros([0,2])
for arm in self.sample_inds_dict:
std_arm = np.std(self.y[self.sample_inds_dict[arm], :], axis= 0, keepdims= True)
std_matrix = np.append(std_matrix, std_arm, axis = 0)
self.std_matrix= std_matrix
def load_diabetes(self, file_name, scale_method):
if scale_method== 'standardize':
diabetes_dict = pickle_read(file_name)
scaler = StandardScaler()
data= diabetes_dict['y']
scaler.fit(data)
data_stand= scaler.transform(data)
self.y = data_stand
self.x= diabetes_dict['x']
self.median_matrix = self.y
elif scale_method== 'no_standardization':
diabetes_dict = pickle_read(file_name)
self.y =diabetes_dict['y']
self.x= diabetes_dict['x']
self.median_matrix = self.y
def find_pessimistic_and_eliminate(arm_dict, M, D):
empirical_median_matrix = np.zeros([0, M])
existing_arm_index = np.zeros([0, ])
U_vec= np.zeros([0, ])
for arm in arm_dict:
empirical_median_vec= np.zeros([1, M])
U= arm_dict[arm]['Ui']
U_vec= np.append(U_vec, [U], axis= 0)
for j in range(M):
empirical_median_vec[0, j] = arm_dict[arm]['mi_hat'][j]
empirical_median_matrix= np.append(empirical_median_matrix, empirical_median_vec, axis= 0)
existing_arm_index= np.append(existing_arm_index, int(arm))
lower_confidence= empirical_median_matrix- np.expand_dims(U_vec, axis=1) - D
pess_ind, non_pess_ind= pareto(lower_confidence, K= lower_confidence.shape[0])
dominated= np.zeros([len(existing_arm_index), 1])
non_dominated= np.ones([len(existing_arm_index), 1])
for i, arm in enumerate(existing_arm_index):
median_vec= empirical_median_matrix[i]
U= U_vec[i]
for _pess_ind in pess_ind:
median_vec_pess= empirical_median_matrix[_pess_ind]
U_pess= U_vec[_pess_ind]
if _pess_ind == i:
pass
elif np.all(median_vec_pess- U_pess -D>= (median_vec + D +U)):
dominated[i, 0] = 1
non_dominated[i, 0] = 0
if len(dominated.nonzero()[0]) == 0:
return None
else:
return existing_arm_index[dominated.nonzero()[0]]