forked from ej0cl6/deep-active-learning
-
Notifications
You must be signed in to change notification settings - Fork 0
/
data.py
141 lines (111 loc) · 4.77 KB
/
data.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
import numpy as np
import torch
from torchvision import datasets
class Data:
def __init__(self, X_train, Y_train, X_test, Y_test, X_pool, Y_pool, handler):
self.X_train = X_train
self.Y_train = Y_train
self.X_test = X_test
self.Y_test = Y_test
self.X_pool = X_pool
self.Y_pool = Y_pool
self.handler = handler
self.n_train = len(X_train)
self.n_pool = len(X_pool)
self.n_test = len(X_test)
self.labeled_idxs = np.zeros(self.n_train, dtype=bool)
self.labeled_pool_idxs = np.zeros(self.n_pool, dtype=bool)
def initialize_labels(self, num):
# generate initial labeled data based on num of initial labeled samples from X_train we decided
tmp_idxs = np.arange(self.n_train)
# shuffle idxs
np.random.shuffle(tmp_idxs)
# label a subset
self.labeled_idxs[tmp_idxs[:num]] = True
def get_labeled_pool_data(self):
return np.arange(self.n_pool)[self.labeled_pool_idxs]
def get_labeled_data(self):
# grab labeled train idxs
labeled_train_idxs = np.arange(self.n_train)[self.labeled_idxs]
X_train = self.X_train[labeled_train_idxs]
Y_train = self.Y_train[labeled_train_idxs]
# grab labeled pool idxs if any
labeled_pool_idxs = self.get_labeled_pool_data()
X_pool = self.X_pool[labeled_pool_idxs]
Y_pool = self.Y_pool[labeled_pool_idxs]
# Combine training and labelled pool set
X_train_combined = torch.cat((X_train, X_pool), 0)
Y_train_combined = torch.cat((Y_train, Y_pool), 0)
Y_train_combined = Y_train_combined.long()
# train model on labeled data
return self.handler(X_train_combined, Y_train_combined)
def get_unlabeled_data(self):
# Get unlabeled data from X_pool
unlabeled_idxs = np.arange(self.n_pool)[~self.labeled_pool_idxs]
return unlabeled_idxs, self.handler(self.X_pool[unlabeled_idxs], self.Y_pool[unlabeled_idxs])
def get_train_data(self):
return self.handler(self.X_train, self.Y_train)
def get_test_data(self):
return self.handler(self.X_test, self.Y_test)
def cal_test_acc(self, preds):
return 1.0 * (self.Y_test==preds).sum().item() / self.n_test
def cal_test_acc_per_class(self, preds):
labels = self.Y_test
preds = preds
acc_per_class = {'0':0, '1':0, '2':0, '3':0, '4':0, '5':0,
'6': 0, '7': 0, '8': 0, '9': 0}
for i in range (10):
acc = 0
total_class_samples = len(labels[labels == i])
for j in range(len(labels)):
if labels[j] == i:
if preds[j] == i:
acc += 1 / total_class_samples
acc_per_class['{}'.format(i)] = acc
return acc_per_class
def train_split(train_data):
'''
Get training data made up of 0-3 digits
'''
# split train set to contain only 0-3 digits
x_train = train_data.data[:40000]
y_train = train_data.targets[:40000]
x_zero = x_train[y_train == 0]
x_one = x_train[y_train == 1]
x_two = x_train[y_train == 2]
x_three = x_train[y_train == 3]
y_zero = torch.from_numpy(np.array([0] * len(x_zero)))
y_one = torch.from_numpy(np.array([1] * len(x_one)))
y_two = torch.from_numpy(np.array([2] * len(x_two)))
y_three = torch.from_numpy(np.array([3] * len(x_three)))
x_train_03 = torch.cat((x_zero, x_one, x_two, x_three), 0)
y_train_03 = torch.cat((y_zero, y_one, y_two, y_three), 0)
return x_train_03, y_train_03
def get_pool_set(train_data):
'''
Get pool set made up of 4-9 training data digits
'''
# split train set to contain only 4-9 digits
x_train = train_data.data[:40000]
y_train = train_data.targets[:40000]
x_four = x_train[y_train == 4]
x_five = x_train[y_train == 5]
x_six = x_train[y_train == 6]
x_seven = x_train[y_train == 7]
x_eight = x_train[y_train == 8]
x_nine = x_train[y_train == 9]
labels = np.array([])
for i in range(4,10):
labels = np.append(labels,np.array([i] * len(x_train[y_train == i])))
y_train_49 = torch.from_numpy(labels)
x_train_49 = torch.cat((x_four, x_five, x_six, x_seven,
x_eight, x_nine), 0)
return x_train_49, y_train_49
def get_MNIST(handler):
raw_train = datasets.MNIST('./data/MNIST', train=True, download=True)
raw_test = datasets.MNIST('./data/MNIST', train=False, download=True)
# grab training set of digits 0-3
x_train, y_train = train_split(raw_train)
# grab pool set of digits 4-9
x_pool, y_pool = get_pool_set(raw_train)
return Data(x_train, y_train, raw_test.data[:40000], raw_test.targets[:40000], x_pool, y_pool, handler)