-
Notifications
You must be signed in to change notification settings - Fork 28
/
parse_mturk.py
199 lines (165 loc) · 7.42 KB
/
parse_mturk.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
import csv
import sys
from collections import defaultdict
import numpy as np
csv_file = sys.argv[1]
header=None
allRows=[]
totalCorrect=0 #^666 66666666666666666666666666666666 666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666 666n6 66666 66 6 6 6 6 6 66 6 66 6 6b6666666666666666666666666 6 n6 6 6 66 6 6 6 6 66 66 6 6 66 6 6666
totalGuessRight=0
noGoldGuessRight=0
goldRight=0
totalGold=0
#workerStats=defaultdict(lambda : {'guessRight': 0, 'noGoldGuessRIght':0, 'total': 0, 'correct': 0, 'goldRight':0, 'goldTotal':0, 'eGen':0, 'eReal':0})
workerStats=defaultdict(lambda : defaultdict(lambda:0))
total=0
annom=0
with open(csv_file) as f:
reader = csv.reader(f, delimiter=',', quotechar='"')
for row in reader:
if header is None:
header=row
continue
human = row[33]=='true'
generated = row[32]=='true'
#assert(human != generated)
gt_human = row[31]=='real'
gold_std = row[31]=='gold'
if len(row[34])>0:
correct = row[34]=='false'
else:
correct = row[35]=='true'
workerId = row[15]
total+=1
approve=True
failedGold=False
failedTrans=False
workerStats[workerId]['total']+=1
if gt_human==human and human!=generated:
totalGuessRight+=1
workerStats[workerId]['guessRight']+=1
if not gold_std:
noGoldGuessRight+=1
workerStats[workerId]['noGoldGuessRight']+=1
if human:
workerStats[workerId]['noGoldRightGuessHuman']+=1
else:
workerStats[workerId]['noGoldRightGuessGen']+=1
elif gt_human and not human and not gold_std:
workerStats[workerId]['noGoldWrongGuessGen']+=1
elif not gt_human and human and not gold_std:
workerStats[workerId]['noGoldWrongGuessHuman']+=1
if human==generated:
annom+=1
if correct:
totalCorrect+=1
workerStats[workerId]['correct']+=1
else:
approve=False
failedTrans=True
if gold_std:
totalGold+=1
workerStats[workerId]['goldTotal']+=1
if generated:
goldRight+=1
workerStats[workerId]['goldRight']+=1
else:
approve=False
failedGold=True
if gt_human:
workerStats[workerId]['eReal']+=1
else:
workerStats[workerId]['eGen']+=1
#allRows.append((row,[workerId,accept
if approve:
row.append('Approve')
elif failedGold:
row+=['','Misclassified obvious image (diliberately poor quality)']
elif failedTrans:
row+=['','Failed attention task (selected incorrect transcription)']
allRows.append(row)
with open('app_rej.csv','w') as out:
csvwriter = csv.writer(out, delimiter=',',
quotechar='"', quoting=csv.QUOTE_MINIMAL)
csvwriter.writerow(header)
for row in allRows:
csvwriter.writerow(row)
print('Total')
print('Correct: {}, guess right: {}, gold: {}/{}'.format(totalCorrect/total,totalGuessRight/total,goldRight,totalGold))
print('annom {}/{}'.format(annom,total))
#print(workerStats)
filteredWorkers=0
filteredRightGuessHuman=0.0
filteredWrongGuessHuman=0.0
filteredRightGuessGen=0.0
filteredWrongGuessGen=0.0
keptG=0
keptT=0
reG=0
reT=0
suspect=[]
accWrkr=[]
accWrkrId=[]
for workerId,stats in workerStats.items():
print('{}\t{}\t{}\t{}\tcorrect:{}\tright:{}'.format(workerId,stats['total'],workerStats[workerId]['eReal'],workerStats[workerId]['eGen'],stats['correct']/stats['total'],stats['guessRight']/stats['total']))
if stats['goldTotal']>0:
print('\t\tgold:{}/{}'.format(stats['goldRight'],stats['goldTotal']))
goldRatio = stats['goldRight']/stats['goldTotal']
else:
goldRatio = 1
correctRatio = stats['correct']/stats['total']
#if stats['guessRight']/stats['total']>0.5:
#if stats['total']-stats['goldTotal']>0 and (stats['noGoldRightGuessHuman']+stats['noGoldRightGuessGen'])/(stats['total']-stats['goldTotal'])>0.5:
#if correctRatio>0.9 and goldRatio>0.95 and stats['total']>5:
if correctRatio>0.9 and stats['total']>5: #final crit
filteredWorkers+=1
filteredRightGuessHuman+=stats['noGoldRightGuessHuman']
filteredWrongGuessHuman+=stats['noGoldWrongGuessHuman']
filteredRightGuessGen+=stats['noGoldRightGuessGen']
filteredWrongGuessGen+=stats['noGoldWrongGuessGen']
accWrkr.append( (stats['noGoldRightGuessGen']+stats['noGoldRightGuessHuman'])/((stats['noGoldRightGuessGen']+stats['noGoldRightGuessHuman']+stats['noGoldWrongGuessGen']+stats['noGoldWrongGuessHuman'])))
accWrkrId.append(workerId)
keptG+=stats['goldRight']
keptT+=stats['goldTotal']
else:#if stats['total']<3:
reG+=stats['goldRight']
reT+=stats['goldTotal']
if correctRatio<0.75 or (stats['goldTotal']>4 and goldRatio<0.5):
suspect.append((workerId,stats))
filteredTotal = filteredRightGuessHuman+filteredWrongGuessHuman+filteredRightGuessGen+filteredWrongGuessGen
filteredRight = filteredRightGuessHuman+filteredRightGuessGen
print('kept instances:{}/{} {:.3}, workers: {:.3}\twere right: {:.3}'.format(filteredTotal,total,filteredTotal/total,filteredWorkers/len(workerStats),filteredRight/filteredTotal))
print(' Guessed: human computer')
print('Actually human: {:.3} {:.3}'.format(filteredRightGuessHuman/filteredTotal,filteredWrongGuessGen/filteredTotal))
print('Actually generated: {:.3} {:.3}'.format(filteredWrongGuessHuman/filteredTotal,filteredRightGuessGen/filteredTotal))
print('worker std dev: {}'.format(np.std(accWrkr)))
accWrkr=list(zip(accWrkr,accWrkrId))
accWrkr.sort(reverse=True,key=lambda a:a[0])
bestWrkr = accWrkr[:len(accWrkr)//10]
#print(bestWrkr)
bestScore,bestIds = zip(*bestWrkr)
filteredRightGuessHuman=0.0
filteredWrongGuessHuman=0.0
filteredRightGuessGen=0.0
filteredWrongGuessGen=0.0
#print(bestIds)
for workerId,stats in workerStats.items():
if workerId in bestIds:
filteredRightGuessHuman+=stats['noGoldRightGuessHuman']
filteredWrongGuessHuman+=stats['noGoldWrongGuessHuman']
filteredRightGuessGen+=stats['noGoldRightGuessGen']
filteredWrongGuessGen+=stats['noGoldWrongGuessGen']
filteredTotal = filteredRightGuessHuman+filteredWrongGuessHuman+filteredRightGuessGen+filteredWrongGuessGen
filteredRight = filteredRightGuessHuman+filteredRightGuessGen
print('top 10% accuracy: {}'.format(filteredRight/filteredTotal))
print('Actually human: {:.3} {:.3}'.format(filteredRightGuessHuman/filteredTotal,filteredWrongGuessGen/filteredTotal))
print('Actually generated: {:.3} {:.3}'.format(filteredWrongGuessHuman/filteredTotal,filteredRightGuessGen/filteredTotal))
print('Top 10% GT Gen: {}, GT human: {}'.format(filteredRightGuessHuman+filteredWrongGuessGen,filteredWrongGuessHuman+filteredRightGuessGen))
print(keptG/keptT)
print(reG/reT)
#print('suspect')
#for workerId,stats in suspect:
# print('{}\t{}\t{}\t{}\tcorrect:{}\tright:{}'.format(workerId,stats['total'],workerStats[workerId]['eReal'],workerStats[workerId]['eGen'],stats['correct']/stats['total'],stats['guessRight']/stats['total']))
# if stats['goldTotal']>0:
# print('\t\tgold:{}/{}'.format(stats['goldRight'],stats['goldTotal']))
#print(header)