-
Notifications
You must be signed in to change notification settings - Fork 3
/
model_results_binary.py
196 lines (155 loc) · 6.77 KB
/
model_results_binary.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
from argparse import ArgumentParser
import pandas as pd
from sklearn.preprocessing import LabelEncoder
import utils
from model_results import print_metric, print_single_metric
class ModelPerformance:
"""Summary of binary model performance metrics."""
precision_majority_all: float
recall_majority_all: float
f1_majority_all: float
precision_unanimous: float
recall_unanimous: float
f1_unanimous: float
precision_majority_controversial: float
recall_majority_controversial: float
f1_majority_controversial: float
def print(self):
print("--- Unanimous + controversial: majority label taken as real label ---")
print_single_metric("Precision:", self.precision_majority_all)
print_single_metric("Recall:", self.recall_majority_all)
print_single_metric("F1:", self.f1_majority_all)
print()
print("--- Unanimous only: unanimous label taken as real label ---")
print_single_metric("Precision:", self.precision_unanimous)
print_single_metric("Recall:", self.recall_unanimous)
print_single_metric("F1:", self.f1_unanimous)
print()
print("--- Controversial only: majority label taken as real label ---")
print_single_metric("Precision:", self.precision_majority_controversial)
print_single_metric("Recall:", self.recall_majority_controversial)
print_single_metric("F1:", self.f1_majority_controversial)
class Comparison:
perf1: ModelPerformance
perf2: ModelPerformance
def __init__(self,
perf1: ModelPerformance,
perf2: ModelPerformance):
self.perf1 = perf1
self.perf2 = perf2
def print(self):
print("--- Unanimous + controversial: majority label taken as real label ---")
print_metric("Precision:", self.perf1.precision_majority_all, self.perf2.precision_majority_all)
print_metric("Recall:", self.perf1.recall_majority_all, self.perf2.recall_majority_all)
print_metric("F1:", self.perf1.f1_majority_all, self.perf2.f1_majority_all)
print()
print("--- Unanimous only: unanimous label taken as real label ---")
print_metric("Precision:", self.perf1.precision_unanimous, self.perf2.precision_unanimous)
print_metric("Recall:", self.perf1.recall_unanimous, self.perf2.recall_unanimous)
print_metric("F1:", self.perf1.f1_unanimous, self.perf2.f1_unanimous)
print()
print("--- Controversial only: majority label taken as real label ---")
print_metric("Precision:", self.perf1.precision_majority_controversial,
self.perf2.precision_majority_controversial)
print_metric("Recall:", self.perf1.recall_majority_controversial, self.perf2.recall_majority_controversial)
print_metric("F1:", self.perf1.f1_majority_controversial, self.perf2.f1_majority_controversial)
def compare(model1: tuple[list[list[int]], list[bool]],
model2: tuple[list[list[int]], list[bool]]) -> Comparison:
perf1 = performance(model1[0], model1[1])
perf2 = performance(model2[0], model2[1])
return Comparison(perf1, perf2)
def performance(real: list[list[int]], pred: list[bool]) -> ModelPerformance:
"""
Calculates binary model performance based on true labels (voted) and predicted labels.
:param real: True labels, as votes -- each element of the list is a list of votes
:param pred: Predicted labels
:return: Performance metrics
"""
tp_all = 0
fp_all = 0
fn_all = 0
tp_unan = 0
fp_unan = 0
fn_unan = 0
tp_contr = 0
fp_contr = 0
fn_contr = 0
assert len(real) == len(pred)
for (i, label_set) in enumerate(real):
real_votes_positive = sum(label_set)
predicted_label = pred[i]
unan_positive = real_votes_positive == len(label_set)
if real_votes_positive == 0 or unan_positive: # Unanimous
if predicted_label:
if unan_positive:
tp_all += 1
tp_unan += 1
else:
fp_all += 1
fp_unan += 1
else:
if unan_positive:
fn_all += 1
fn_unan += 1
elif real_votes_positive * 2 != len(label_set): # Controversial but a majority exists
majority_positive = real_votes_positive * 2 > len(label_set)
if predicted_label:
if majority_positive:
tp_all += 1
tp_contr += 1
else:
fp_all += 1
fp_contr += 1
else:
if majority_positive:
fn_all += 1
fn_contr += 1
perf = ModelPerformance()
perf.precision_majority_all = tp_all / (tp_all + fp_all)
perf.recall_majority_all = tp_all / (tp_all + fn_all)
perf.f1_majority_all = 2 * tp_all / (2 * tp_all + fp_all + fn_all)
if tp_unan + fp_unan > 0:
perf.precision_unanimous = tp_unan / (tp_unan + fp_unan)
perf.recall_unanimous = tp_unan / (tp_unan + fn_unan)
perf.f1_unanimous = 2 * tp_unan / (2 * tp_unan + fp_unan + fn_unan)
else:
perf.precision_unanimous = float("NaN")
perf.recall_unanimous = float("NaN")
perf.f1_unanimous = float("NaN")
if tp_contr + fp_contr > 0:
perf.precision_majority_controversial = tp_contr / (tp_contr + fp_contr)
perf.recall_majority_controversial = tp_contr / (tp_contr + fn_contr)
perf.f1_majority_controversial = 2 * tp_contr / (2 * tp_contr + fp_contr + fn_contr)
else:
perf.precision_majority_controversial = float("NaN")
perf.recall_majority_controversial = float("NaN")
perf.f1_majority_controversial = float("NaN")
return perf
def get_y_from_predicted(pq_path: str) -> list[bool]:
df = pd.read_parquet(pq_path)
return df["decision"].tolist()
def get_y_from_true(pq_path: str) -> list[list[int]]:
df = pd.read_parquet(pq_path)
labels = df["labels"].apply(list).tolist()
label_enc = LabelEncoder()
label_enc.fit(utils.flatten(labels))
assert len(label_enc.classes_) == 2
labels_bin = [label_enc.transform(la).tolist() for la in labels]
return labels_bin
def main() -> None:
parser = ArgumentParser()
parser.add_argument("true_y")
parser.add_argument("model1_pred")
parser.add_argument("model2_pred", nargs="?", default=None)
args = parser.parse_args()
true_y = get_y_from_true(args.true_y)
pred1 = get_y_from_predicted(args.model1_pred)
if args.model2_pred:
pred2 = get_y_from_predicted(args.model2_pred)
comp = compare((true_y, pred1), (true_y, pred2))
comp.print()
else:
perf = performance(true_y, pred1)
perf.print()
if __name__ == "__main__":
main()