-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathevaluate.py
270 lines (225 loc) · 9.82 KB
/
evaluate.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
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
'''
* @author Waldinsamkeit
* @email [email protected]
* @create date 2020-10-26 16:36:19
* @desc
'''
import itertools
from typing import Any, Dict, Optional, Sequence, Tuple
from numpy.lib.scimath import sqrt
import sklearn
from sklearn.metrics import confusion_matrix
import numpy as np
from scipy.special import comb
class EvalUnit(object):
"""Smalled Evaluating metrics unit
Attribute:
tp: True Positive item
fp: False Positive item
fn: False Negative item
tn: True Negative item
name: A label to describe specific instance
"""
def __init__(self, tn:int=0, fp:int=0, fn:int=0, tp:int=0, name:Optional[int]=None) -> None:
super(EvalUnit,self).__init__()
self.name = name if name is not None else "None"
self.tp = tp
self.fp = fp
self.fn = fn
self.tn = tn
def __repr__(self):
desc = '\n--------- Desc EvalUnit {}---------\n'.format(self.name)
desc += 'True Positive:{} \nTrue Negative:{} \nFalse Positive:{} \nFalse Negative:{} \n'.format(
self.tp, self.tn, self.fp, self.fn
)
desc += 'Accuracy:{:.4f} \nPrecision:{:.4f} \nRecall:{:.4f} \nF1-Score:{:.4f}'.format(
self.accuracy(), self.precision(), self.recall(), self.f1_score()
)
return desc
def __add__(self,other) -> "EvalUnit":
return EvalUnit(
self.tn + other.tn,
self.fp + other.fp,
self.fn + other.fn,
self.tp + other.tp,
)
def __iadd__(self,other) -> "EvalUnit":
self.tn += other.tn
self.fp += other.fp
self.fn += other.fn
self.tp += other.tp
return self
def accuracy(self) -> float:
return float(self.tn + self. tp) / (self.fp + self.fn + self.tn + self.tp) if (self.fp + self.fn + self.tn + self.tp) !=0 else 0.
def f1_score(self) -> float:
r = self.recall()
p = self.precision()
return 2 * r * p / (p + r) if p + r != 0 else 0.
def precision(self) -> float:
return float(self.tp) / (self.tp + self.fp) if (self.tp + self.fp) != 0 else 0.
def recall(self) -> float:
return float(self.tp) / (self.tp + self.fn) if (self.tp + self.fn) != 0 else 0.
def metrics(self) -> Tuple[float]:
return (self.accuracy(), self.precision(), self.recall(), self.f1_score())
def metrics2dict(self) -> Dict:
return {
"Accuracy" : self.accuracy(),
"Precision" : self.precision(),
"Recall" : self.recall(),
"F1_score" : self.f1_score()
}
def binary_confusion_matrix_evaluate(y_true:Sequence[Any], y_pred:Sequence[Any]) -> EvalUnit:
# import pdb; pdb.set_trace()
tn, fp, fn, tp = confusion_matrix(y_true,y_pred,labels = [0,1]).ravel()
return EvalUnit(tn,fp,fn,tp)
""" ---------------- cluster Metrics ---------------- """
"""
I reimplemet adjusted_rand_index, fowlkes_mallows_scores.
In order to understand algoritmn
But In reality, We can call related API directly from sklearn.
"""
''' helper function'''
def helper_trans_to_element2clusterid(cluster:Dict) -> Dict:
ele2cluster = {}
for key,value in cluster.items():
if key not in ele2cluster:
ele2cluster[key] = []
ele2cluster[key].append(value)
return ele2cluster
''' '''
def cluster_confusion_matrix(pred_cluster:Dict, target_cluster:Dict) -> EvalUnit:
""" simulate confusion matrix
Args:
pred_cluster: Dict element: cluster_id （cluster_id from 0 to max_size）| predicted clusters
target_cluster: Dict element:cluster_id （cluster_id from 0 to max_size) | target clusters
Returns:
In order to return detailed data, It will return a EvalUnit,
"""
pred_elements = list(pred_cluster.keys())
target_elements = list(target_cluster.keys())
it = itertools.product(pred_elements,target_elements)
tp,fp,tn,fn = 0,0,0,0
for x,y in it:
if x != y:#other word
x_cluster = pred_elements[x]
x_cluster_ = target_elements[x]
y_cluster = pred_elements[y]
y_cluster_ = target_elements[y]
if x_cluster == y_cluster and x_cluster_ == y_cluster_:
tp += 1
elif x_cluster != y_cluster and x_cluster_ != y_cluster_:
tn += 1
elif x_cluster == y_cluster and x_cluster_ != y_cluster_:
fp += 1
else:
fn +=1
return EvalUnit(tp,tn,fp,fn,'rand_index')
def get_rand_index(unit:EvalUnit) -> float:
return unit.precision
def get_fowlkes_mallows_score(unit:EvalUnit) -> float:
FMI = unit.tp / sqrt((unit.tp + unit.fp) * (unit.tp+ unit.fn))
def fowlkes_mallows_score(pred_cluster: Dict, target_cluster: Dict) -> float:
unit = cluster_confusion_matrix(pred_cluster,target_cluster)
return get_fowlkes_mallows_score(unit)
def rand_index(pred_cluster: Dict, target_cluster: Dict) -> float:
"""Use contingency_table to get RI directly
RI = Accuracy = (TP+TN)/(TP,TN,FP,FN)
Args:
pred_cluster: Dict element:cluster_id （cluster_id from 0 to max_size）| predicted clusters
target_cluster: Dict element:cluster_id （cluster_id from 0 to max_size) | target clusters
Return:
RI (float)
"""
pred_cluster_ = helper_trans_to_element2clusterid(pred_cluster)
target_cluster_ = helper_trans_to_element2clusterid(target_cluster)
pred_cluster_size = len(pred_cluster_)
target_cluster_size = len(target_cluster_)
contingency_table = np.zeros((pred_cluster_size,target_cluster_size))
for i, p_cluster in enumerate(pred_cluster_):
for j, t_cluster in enumerate(target_cluster_):
#find common element
l = [*p_cluster,*t_cluster]
contingency_table[i][j] = len(l) - len(set(l))
s = comb(np.sum(contingency_table), 2)
a = 0
for i in np.nditer(contingency_table):
a += comb(i,2)
return a/s
def adjusted_rand_index(pred_cluster:Dict, target_cluster:Dict):
"""Docstring
Using Contingency Matrix to calculate ARI
Continggency Matrix
--------------------------------
XY | Y_1 Y_2 ... Y_s | sums
--------------------------------
X_1 | n_11 n_12 ... n_1s | a_1
X_2 | n_21 n_22 ... n_2s | a_2
... | ... ... ... ... | ...
X_r | n_r1 n_r2 ... n_rs | a_r
sum | b_1 b_2 ... b_s |
--------------------------------
f(x) = comb(x,2)
ARI = [ sum f(n_ij) - sum f(a_ij) * sum f(b_ij) / f(n) ] /
[0.5 * [ sum f(a_ij) + sum f(b_ij)] - sum f(a_ij) * sum f(b_ij) / f(n)]
Args:
pred_cluster: Dict cluster_id: List[element] （cluster_id from 0 to max_size）| predicted clusters
target_cluster: Dict cluster_id: List[element] （cluster_id from 0 to max_size) | target clusters
Return:
ARI (float)
"""
pred_cluster_ = helper_trans_to_element2clusterid(pred_cluster)
target_cluster_ = helper_trans_to_element2clusterid(target_cluster)
pred_cluster_size = len(pred_cluster_)
target_cluster_size = len(target_cluster_)
contingency_table = np.zeros((pred_cluster_size, target_cluster_size))
for i, p_cluster in enumerate(pred_cluster_):
for j, t_cluster in enumerate(target_cluster_):
#find common element
l = [*p_cluster,*t_cluster]
contingency_table[i][j] = len(l) - len(set(l))
s = comb(np.sum(contingency_table), 2)
ij = 0
for i in np.npiter(contingency_table):
ij += comb(i,2)
pred_sum = np.sum(contingency_table, axis=1)
target_sum = np.sum(contingency_table, aixs=0)
pred_comb_sum = 0
for i in np.npiter(pred_sum):
pred_comb_sum += comb(i,2)
target_comb_sum = 0
for i in np.npiter(target_sum):
target_comb_sum += comb(i,2)
tmp = pred_comb_sum * target_comb_sum / s
ARI = (ij - tmp) / (0.5*(pred_comb_sum+target_comb_sum) - tmp)
return ARI
"""
Below function is inference of sklearn
I changed the input data type slightly
"""
'''helper function'''
def helper_trans_to_labelsequence(cluster:Dict,cluster_:Dict)-> Any:
keys = cluster.keys()
label_sequence = []
label_sequence_ = []
for element in keys:
label_sequence.append(cluster[element])
label_sequence_.append(cluster_[element])
return np.array(label_sequence), np.array(label_sequence_)
def metrics_adjusted_randn_index(pred_cluster:Dict, target_cluster:Dict) -> Any:
pred_sequence,target_sequence = helper_trans_to_labelsequence(pred_cluster,target_cluster)
return 'ARI', sklearn.metrics.adjusted_rand_score(labels_true = target_sequence, labels_pred = pred_sequence)
def metrics_normalized_mutual_info_score(pred_cluster:Dict, target_cluster:Dict) -> Any:
pred_sequence,target_sequence = helper_trans_to_labelsequence(pred_cluster,target_cluster)
return 'NMI', sklearn.metrics.normalized_mutual_info_score(labels_true = target_sequence, labels_pred = pred_sequence)
def metrics_fowlkes_mallows_score(pred_cluster:Dict, target_cluster:Dict) ->Any:
pred_sequence,target_sequence = helper_trans_to_labelsequence(pred_cluster,target_cluster)
return 'FMI', sklearn.metrics.fowlkes_mallows_score(labels_true = target_sequence, labels_pred = pred_sequence)
def cluster_metrics_eval(pred_cluster:Dict, target_cluster:Dict)->Sequence[Any]:
pred_sequence,target_sequence = helper_trans_to_labelsequence(pred_cluster,target_cluster)
return {
('ARI', sklearn.metrics.adjusted_rand_score(labels_true = target_sequence, labels_pred = pred_sequence)),
('NMI', sklearn.metrics.normalized_mutual_info_score(labels_true = target_sequence, labels_pred = pred_sequence)),
('FMI', sklearn.metrics.fowlkes_mallows_score(labels_true = target_sequence, labels_pred = pred_sequence))
}
if __name__ == '__main__':
print(1)