forked from matthewfl/nlp-entity-convnet
/
evaluation.py
218 lines (200 loc) · 7.81 KB
/
evaluation.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
def evalCurrentState(queries, trainingData=True, numSamples=50000):
all_measured = 0
all_correct = 0
all_trained = 0
all_wrong = 0
for qu in queries.values():
for en in qu.values():
if en['training'] != trainingData:
continue
if en['gold']:
if all_trained > numSamples:
break
all_measured += 1
all_trained += len(en['vals'].values())
sv = sorted(en['vals'].items(), key=lambda x: x[1])
#m = max(en['vals'].values())
if sv[-1][0] in en['gold']:
all_correct += 1
else:
if len(set(en['gold']) & set(en['vals'].keys())) != 0:
# print sv[-1][0], en['gold']
all_wrong += 1
# for g in en['gold']:
# if en['vals'].get(en['gold'][0]) == m and m != 0:
# all_correct += 1
# break
r = all_measured, all_wrong, float(all_correct) / all_measured
print r
return r
def evalCurrentStateRank(queries, trainingData=True, numSamples=50000):
all_measured = 0
all_correct_place = 0
p_counts = dict((k,0) for k in range(0,10))
all_trained = 0
for qu in queries.values():
for en in qu.values():
if en['training'] != trainingData:
continue
if en['gold']:
if all_trained > numSamples:
break
svals = sorted(en['vals'].values(), key=lambda x: 0 if not isinstance(x, tuple) else -x[0])
gv = en['vals'][en['gold']]
if gv == 0:
continue
all_measured += 1
for i in xrange(len(svals)):
if svals[i] == gv:
if i < 10:
p_counts[i] += 1
all_correct_place += i + 1
break
r = all_measured, float(all_correct_place) / all_measured, p_counts
print r
return r
def evalCurrentStateF1(queries, trainingData=True, numSamples=50000):
correct = 0
precDenom = 0
recDenom = 0
all_trained = 0
all_measured = 0
for qu in queries.values():
if qu.values()[0]['training'] != trainingData:
continue
allGold = set()
allChoosen = set()
if all_trained > numSamples:
break
for en in qu.values():
if en['gold']: # we can eval this item
all_measured += 1
allGold.add(en['gold'])
svals = sorted(en['vals'].values())
picked = None
for k,v in en['vals'].iteritems():
all_trained += 1
if v == svals[-1]:
picked = k
allChoosen.add(picked)
#if svals[0] == svals[1] and en['gold'] != picked:
# raise NotImplementedError()
# if en['gold'] == picked:
# correct += 1
# if len(svals) > 5 and en['gold'] != picked:
# raise NotImplementedError()
precDenom += len(allChoosen)
recDenom += len(allGold)
correct += len(allGold & allChoosen)
correct = float(correct)
prec = correct / precDenom
rec = correct / recDenom
f1 = 2 * prec * rec / (prec + rec)
r = all_measured, 'Prec = {}/{} = {}, Rec = {}/{} = {}, F1 = {}'.format(
correct, precDenom, prec,
correct, recDenom, rec,
f1)
print r
return r
from collections import defaultdict
def evalCurrentStateFahrni(queries, trainingData=True, numSamples=50000):
def renderF1(corr, precDenom, recDenom):
prec = float(corr) / precDenom
rec = float(corr) / recDenom
return 'Prec = {}/{} = {}, Rec = {}/{} = {}, F1 = {}'.format(
corr, precDenom, prec,
corr, recDenom, rec,
2 * prec * rec / (prec + rec)
)
counter = defaultdict(lambda: 0)
all_trained = 0
for qu in queries.values():
if qu.values()[0]['training'] != trainingData:
continue
for en in qu.values():
if en['gold']:
if all_trained > numSamples:
continue
gold = en['gold']
svals = sorted(en['vals'].items(), key=lambda x: x[1])
picked = svals[-1][0]
all_trained += len(svals)
label = None
if len(gold) == 1 and gold[0] == '-NIL-':
if picked == '-NIL-':
label = 'cNIL'
else:
label = 'wNIL_KB'
elif picked in gold:
label = 'cKB'
elif picked == '-NIL-':
label = 'wKB_NIL'
else:
label = 'wKB_KB'
counter[label] += 1
rr = 'KB: {}'.format(renderF1(counter['cKB'], counter['cKB'] + counter['wKB_KB'] + counter['wNIL_KB'], counter['cKB'] + counter['wKB_KB'] + counter['wKB_NIL']))
print rr
if counter['cNIL']:
rr2 = 'NIL: {}'.format(renderF1(counter['cNIL'], counter['cNIL'] + counter['wKB_NIL'], counter['cNIL'] + counter['wNIL_KB']))
print rr2
rr += '; '+rr2
# print 'KB:', renderF1(counter['cKB'], counter['cKB'] + counter['wKB_KB'] + counter['wNIL_KB'], counter['cKB'] + counter['wKB_KB'] + counter['wKB_NIL'])
# if counter['cNIL']:
# print 'NIL:', renderF1(counter['cNIL'], counter['cNIL'] + counter['wKB_NIL'], counter['cNIL'] + counter['wNIL_KB'])
return counter, rr
def findWrongItems(queries, trainingData=True, numSamples=50):
# theano overrides map if imported with *
#from __builtin__ import map
ret = []
# surface = set()
for qu in queries.values():
for ek, en in qu.items():
if en['training'] != trainingData:
continue
# for e in en:
if en['gold']:
if len(ret) > numSamples:
return ret
if True:#ek not in surface:
sv = sorted(en['vals'].items(), key=lambda x: x[1])
if sv[-1][0] not in en['gold'] and len(set(en['gold']) & set(en['vals'].keys())) != 0:
# if 'Slayer (Buffy the Vampire Slayer)' in en['gold']:
# got this wrong
ret.append({
'gold': en['gold'],
'ordered': [(s[0], s[1][0], [(' '.join(map(str, a)), len(a)) for a in s[1][1]]) for s in sv][::-1],
'text': ek,
'training': en['training'],
})
# m = max(en['vals'].values())
# g = en['vals'].get(en['gold'][0], 0)
# if g != m and g != 0:
# ret[ek] = en
return ret
def evalNumPossible(queries, qtype=(False,True)):
total = 0
possible = 0
for qu in queries.values():
for q in qu.values():
if q['training'] in qtype:
total += 1
if len(set(q['gold']) & set(q['vals'].keys())) != 0:
possible += 1
return float(possible) / total
def findItms(queries, key):
ret = []
from __builtin__ import map
for qu in queries.values():
for en in qu.values():
if en['training'] == True:
continue
ad = False
for k in en['vals'].keys():
if key in k:
ad = True
if ad:
for k, v in en['vals'].iteritems():
ret.append((k, [(' '.join(map(str, a)), len(a)) for a in v[1]]))
# return ret
print len(ret)
return ret