-
Notifications
You must be signed in to change notification settings - Fork 0
/
comparison.py
201 lines (167 loc) · 7.46 KB
/
comparison.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
import pandas as pd
from collections import OrderedDict
import matplotlib.pyplot as plt
import os
import utils.resources as res
OUTPUT_PATH = os.path.join(
res.output,
'stats')
# TODO: overlap lemmatised/not lemmatised -> examples
# TODO: lexical difference in preds
# TODO: duplicates in hypothesis parts, daganlevy
# poland was divided among russia 4690,4959
# jupiter is big as the earth 1989, 13983, 14006
def size(dataset):
return len(dataset)
def positives(dataset):
return dataset[dataset['entailment'] == True]
def negatives(dataset):
return dataset[dataset['entailment'] == False]
def templates(dataset):
return dataset.text.append(dataset.hypothesis).rename('templates')
def predicates(dataset):
return dataset.tpred.append(dataset.hpred).rename('predicates')
def attributes(dataset):
return dataset.tx.append([
dataset.ty,
dataset.hx,
dataset.hy,
]).rename('attributes')
def unique_templates(dataset):
return set(templates(dataset))
def unique_predicates(dataset):
return set(predicates(dataset))
def unique_attributes(dataset):
return set(attributes(dataset))
def shared_templates(datasetA, datasetB):
templatesA = unique_templates(datasetA)
templatesB = unique_templates(datasetB)
return templatesA.intersection(templatesB)
def shared_predicates(datasetA, datasetB):
predicatesA = unique_predicates(datasetA)
predicatesB = unique_predicates(datasetB)
return predicatesA.intersection(predicatesB)
def shared_predicates_only(datasetA, datasetB):
templatesB = unique_predicates(datasetB)
shared_text = [text in templatesB for text in datasetA.tpred]
shared_hypothesis = [hypothesis in templatesB for hypothesis in datasetA.hpred]
shared_temps = [max(t,h) for t,h in zip(shared_text,shared_hypothesis)]
return datasetA[shared_temps]
def shared_predicates_sorted(datasetA, datasetB):
predsA = predicates(datasetA).value_counts().to_frame()
predsB = predicates(datasetB).value_counts().to_frame()
predsA['rank'] = range(1, predsA.size + 1)
predsB['rank'] = range(1, predsB.size + 1)
C = predsA.join(predsB, lsuffix = 'A', rsuffix = 'B')
C.dropna(inplace = True)
C['avgrank'] = (C['rankA'] + C['rankB']) / 2
C['freq'] = C['predicatesA'] + C['predicatesB']
C.sort_values('avgrank', inplace = True)
return C[['avgrank', 'freq']]
def shared_attributes(datasetA, datasetB):
attributesA = unique_attributes(datasetA)
attributesB = unique_attributes(datasetB)
return attributesA.intersection(attributesB)
def coverage(datasetA, datasetB, coverage_function):
A = coverage_function(datasetA)
B = coverage_function(datasetB)
return len(A & B) / len(B)
def coverage_templates(datasetA, datasetB):
templatesA = unique_templates(datasetA)
templatesB = unique_templates(datasetB)
return len(templatesA & templatesB) / len(templatesB)
def coverage_predicates(datasetA, datasetB):
predsA = unique_predicates(datasetA)
predsB = unique_predicates(datasetB)
return len(predsA & predsB) / len(predsB)
def coverage_attributes(datasetA, datasetB):
attsA = unique_attributes(datasetA)
attsB = unique_attributes(datasetB)
return len(attsA & attsB) / len(attsB)
def jaccard_index(listA, listB):
A = set(listA)
B = set(listB)
union = A.union(B)
intersection = A.intersection(B)
return len(intersection) / len(union)
def dataset_stats(dataset):
return pd.Series(OrderedDict({
'size': size(dataset),
'positives': len(positives(dataset)),
'negatives': len(negatives(dataset)),
'pn_rate': len(positives(dataset)) / len(negatives(dataset)),
'unique_templates': len(unique_templates(dataset)),
'unique_templates_T': dataset['text'].nunique(),
'unique_templates_H': dataset['hypothesis'].nunique(),
'unique_predicates': len(unique_predicates(dataset)),
'unique_predicates_T': dataset['tpred'].nunique(),
'unique_predicates_H': dataset['hpred'].nunique(),
'unique_attributes': len(unique_attributes(dataset)),
'unique_attributes_T': dataset['tx'].append(dataset['ty']).nunique(),
'unique_attributes_H': dataset['hx'].append(dataset['hy']).nunique(),
'attribute_predicate_rate': len(unique_attributes(dataset)) / len(unique_predicates(dataset))
}))
def comparison_stats(datasetA, datasetB):
return pd.Series(OrderedDict({
'shared_templates': len(shared_templates(datasetA, datasetB)),
'shared_predicates': len(shared_predicates(datasetA, datasetB)),
'shared_attributes': len(shared_attributes(datasetA, datasetB)),
'coverage_templates': coverage_templates(datasetA, datasetB),
'coverage_predicates': coverage_predicates(datasetA, datasetB),
'coverage_attributes': coverage_attributes(datasetA, datasetB),
'jaccard_templates': jaccard_index(unique_templates(datasetA), unique_templates(datasetB)),
'jaccard_predicates': jaccard_index(unique_predicates(datasetA), unique_predicates(datasetB)),
'jaccard_attributes': jaccard_index(unique_attributes(datasetA), unique_attributes(datasetB))
}))
def descriptives(dataset):
temps = templates(dataset)
preds = predicates(dataset)
attrs = attributes(dataset)
return pd.DataFrame([
temps.value_counts().describe(),
dataset.text.value_counts().describe(),
dataset.hypothesis.value_counts().describe(),
preds.value_counts().describe(),
dataset.tpred.value_counts().describe(),
dataset.hpred.value_counts().describe(),
attrs.value_counts().describe(),
dataset.tx.append(dataset.ty).rename('attributes_text').value_counts().describe(),
dataset.hx.append(dataset.hy).rename('attributes_hypothesis').value_counts().describe()
])
def top10(series):
top = series.value_counts().head(10)
df = pd.DataFrame(
list(zip(top.index, top.values)),
columns = [top.name, 'frequency'])
df.reset_index(drop = True, inplace = True)
return df
def top10s(dataset):
return pd.concat([
top10(templates(dataset)),
top10(predicates(dataset)),
top10(attributes(dataset))
],
axis = 1)
def compare_datasets(datasetA, datasetB, names = ['A', 'B']):
datasets = [datasetA, datasetB]
stats = pd.DataFrame(
[
dataset_stats(datasetA).append(comparison_stats(datasetA, datasetB)),
dataset_stats(datasetB).append(comparison_stats(datasetB, datasetA))
],
index=names)
shared_preds = shared_predicates_sorted(datasetA, datasetB)
for i, name in enumerate(names):
outpath = os.path.join(OUTPUT_PATH, name)
descriptives(datasets[i]).to_csv(outpath + '_descriptives.csv')
top10s(datasets[i]).to_csv(outpath + '_top10.csv')
top10s(shared_predicates_only(datasets[i], datasets[i-1])).to_csv(outpath + '_top10-shared.csv')
suffix = '_'.join(names) + '.csv'
stats.T.to_csv(os.path.join(OUTPUT_PATH, 'dataset_stats-' + suffix))
shared_preds.to_csv(os.path.join(OUTPUT_PATH, 'shared_predicates-' + suffix))
if __name__ == '__main__':
daganlevy = res.load_dataset('daganlevy', 'tidy')
daganlevy_lemmatised = res.load_dataset('daganlevy_lemmatised', 'tidy')
zeichner = res.load_dataset('zeichner', 'tidy')
compare_datasets(daganlevy, zeichner, names = ['daganlevy', 'zeichner'])
compare_datasets(daganlevy_lemmatised, zeichner, names = ['daganlevy_lemmatised', 'zeichner'])