-
Notifications
You must be signed in to change notification settings - Fork 28
/
split_metadata.py
227 lines (208 loc) · 9.34 KB
/
split_metadata.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
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
import sys
from collections import defaultdict
import json
import numpy as np
import copy
from util import *
def split_and_save_data(vm, target_key_type, method="assign", no_link=False):
# Index metadata by date or camera
vm_dict = defaultdict(list)
for v in vm:
k = to_key(v, target_key_type)
if no_link:
if "url_part" in v: del v["url_part"]
if "url_root" in v: del v["url_root"]
vm_dict[k].append(v)
p = "../data/split/"
check_and_create_dir(p)
print("="*40)
print("="*40)
print("Split data by " + target_key_type)
if method == "random":
vm_train, vm_valid, vm_test = split(vm_dict, target_key_type)
save_json(vm_valid, p+"metadata_validation_random_split_by_"+target_key_type+".json")
save_json(vm_test, p+"metadata_test_random_split_by_"+target_key_type+".json")
save_json(vm_train, p+"metadata_train_random_split_by_"+target_key_type+".json")
elif method == "assign":
if target_key_type == "camera":
three_splits = [
{
"train": ["0-0", "0-3", "0-6", "0-8", "0-12", "0-2", "0-5", "0-9", "0-11", "0-14"],
"valid": ["0-4", "0-10"],
"test": ["1-0", "2-0", "2-1", "2-2", "0-1", "0-7", "0-13"]
}, {
"train": ["0-1", "0-4", "0-7", "0-10", "0-13", "0-0", "0-3", "0-6", "0-8", "0-12"],
"valid": ["0-5", "0-11"],
"test": ["1-0", "2-0", "2-1", "2-2", "0-2", "0-9", "0-14"]
}, {
"train": ["0-2", "0-5", "0-9", "0-11", "0-14", "0-1", "0-4", "0-7", "0-10", "0-13"],
"valid": ["0-3", "0-8"],
"test": ["1-0", "2-0", "2-1", "2-2", "0-0", "0-6", "0-12"]
}, {
"train": ["0-0", "0-1", "0-2", "0-3", "0-5", "0-6", "0-11", "0-12", "0-13", "0-14"],
"valid": ["0-7", "0-9"],
"test": ["1-0", "2-0", "2-1", "2-2", "0-4", "0-8", "0-10"]
}, {
"train": ["0-0", "0-1", "0-2", "0-7", "0-8", "0-9", "0-10", "0-12", "0-13", "0-14"],
"valid": ["0-4", "0-6"],
"test": ["1-0", "2-0", "2-1", "2-2", "0-3", "0-5", "0-11"]
}]
for i in range(len(three_splits)):
print("-"*20)
print("Split %d" % i)
s = three_splits[i]
vm_train, vm_valid, vm_test = split(vm_dict, target_key_type,
train_key=s["train"], valid_key=s["valid"], test_key=s["test"])
save_json(vm_valid, p+"metadata_validation_split_"+str(i)+"_by_"+target_key_type+".json")
save_json(vm_test, p+"metadata_test_split_"+str(i)+"_by_"+target_key_type+".json")
save_json(vm_train, p+"metadata_train_split_"+str(i)+"_by_"+target_key_type+".json")
elif target_key_type == "date":
target_keys = list(vm_dict.keys())
target_keys = sorted(target_keys)[::-1]
train_key, valid_key, test_key = divide_list(target_keys, frac_valid=0.07, frac_test=0.35)
vm_train, vm_valid, vm_test = split(vm_dict, target_key_type,
train_key=train_key, valid_key=valid_key, test_key=test_key)
save_json(vm_valid, p+"metadata_validation_split_by_"+target_key_type+".json")
save_json(vm_test, p+"metadata_test_split_by_"+target_key_type+".json")
save_json(vm_train, p+"metadata_train_split_by_"+target_key_type+".json")
print("The data split is saved in: " + p)
def divide_list(target_keys, frac_valid=0.1, frac_test=0.3):
n_keys = len(target_keys)
n_valid = int(n_keys*frac_valid)
n_test = int(n_keys*frac_test)
test_key = target_keys[:n_test]
valid_key = target_keys[n_test:n_valid+n_test]
train_key = target_keys[n_valid+n_test:]
return (train_key, valid_key, test_key)
def split(vm_dict, target_key_type, train_key=None, valid_key=None, test_key=None):
if train_key is None or valid_key is None or test_key is None:
target_keys = list(vm_dict.keys())
np.random.shuffle(target_keys)
# 10% for validation, 30% for testing
train_key, valid_key, test_key = divide_list(target_keys, frac_valid=0.1, frac_test=0.3)
vm_valid = []
vm_test = []
vm_train = []
for d in vm_dict:
if d in valid_key:
vm_valid += vm_dict[d]
elif d in test_key:
vm_test += vm_dict[d]
elif d in train_key:
vm_train += vm_dict[d]
print("\nTraining:")
print_distribution(vm_train, target_key_type=target_key_type)
print("\nValidation:")
print_distribution(vm_valid, target_key_type=target_key_type)
print("\nTesting:")
print_distribution(vm_test, target_key_type=target_key_type)
n = len(vm_valid) + len(vm_test) + len(vm_train)
print("\nSize of validation set: %d (%.2f)" % (len(vm_valid), len(vm_valid)/n))
print("Size of test set: %d (%.2f)" % (len(vm_test), len(vm_test)/n))
print("Size of training set: %d (%.2f)" % (len(vm_train), len(vm_train)/n))
return (vm_train, vm_valid, vm_test)
def print_distribution(vm, target_key_type):
count_vm = defaultdict(lambda: defaultdict(int))
for v in vm:
k = to_key(v, target_key_type)
label = v["label"]
if label == 1:
count_vm[k]["pos"] += 1
elif label == 0:
count_vm[k]["neg"] += 1
for k in count_vm:
count_vm[k]["sum"] = count_vm[k]["pos"] + count_vm[k]["neg"]
count_vm[k]["pos_%"] = np.round(count_vm[k]["pos"] / count_vm[k]["sum"], 2)
print(json.dumps(count_vm, indent=4))
def to_key(v, target_key_type):
# The file name contains information about camera, date, and bounding box
# We can use this as the key of the dataset for separating training, validation, and test sets
key = v["file_name"].split("-")
if target_key_type == "camera":
return "-".join(key[0:2])
elif target_key_type == "date":
return "-".join(key[2:5])
# Aggregate labels from citizens (label_state) and researchers (label_state_admin)
# "label" means the final aggregated label
# "weight" means the confidence of the aggregated label
def aggregate_label(vm, add_weight=True):
vm = copy.deepcopy(vm)
vm_new = []
for i in range(len(vm)):
has_error = False
v = vm[i]
label_state_admin = v["label_state_admin"]
label_state = v["label_state"]
if label_state_admin == 47: # pos (gold standard)
v["label"] = 1
if add_weight: v["weight"] = 1
print("Warning: found gold standards")
elif label_state_admin == 32: # neg (gold standard)
v["label"] = 0
if add_weight: v["weight"] = 1
print("Warning: found gold standards")
elif label_state_admin == 23: # strong pos
v["label"] = 1
if add_weight:
if label_state == 23: # strong pos
v["weight"] = 1 # (1+1)/2
elif label_state == 16: # strong neg
v["weight"] = 0.5 # (1+0)/2
elif label_state == 20: # weak neg
v["weight"] = 0.66 # (1+0.33)/2
elif label_state == 19: # weak pos
v["weight"] = 0.83 # (1+0.66)/2
else: # not determined by citizens
v["weight"] = 0.75
elif label_state_admin == 16: # strong neg
v["label"] = 0
if add_weight:
if label_state == 23: # strong pos
v["weight"] = 0.5 # (1+0)/2
elif label_state == 16: # strong neg
v["weight"] = 1 # (1+1)/2
elif label_state == 20: # weak neg
v["weight"] = 0.83 # (1+0.66)/2
elif label_state == 19: # weak pos
v["weight"] = 0.66 # (1+0.33)/2
else: # not determined by citizens
v["weight"] = 0.75
else: # not determined by researchers
if label_state == 23: # strong pos
v["label"] = 1
if add_weight: v["weight"] = 1
elif label_state == 16: # strong neg
v["label"] = 0
if add_weight: v["weight"] = 1
elif label_state == 20: # weak neg
v["label"] = 0
if add_weight: v["weight"] = 0.66
elif label_state == 19: # weak pos
v["label"] = 1
if add_weight: v["weight"] = 0.66
else:
has_error = True
if has_error or "label" not in v:
print("Error when aggregating label:")
print(v)
else:
vm_new.append(v)
return vm_new
# Split metadata into training, validation, and test sets
def main(argv):
# Check
if len(argv) > 1:
if argv[1] != "confirm":
print("Must confirm by running: python split_metadata.py confirm")
return
else:
print("Must confirm by running: python split_metadata.py confirm")
return
vm = load_json("../data/metadata.json")
vm = aggregate_label(vm)
method = "assign"
no_link = True
split_and_save_data(vm, "date", method=method, no_link=no_link)
split_and_save_data(vm, "camera", method=method, no_link=no_link)
if __name__ == "__main__":
main(sys.argv)