-
Notifications
You must be signed in to change notification settings - Fork 1
/
algorithm.py
199 lines (145 loc) · 7.26 KB
/
algorithm.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 numpy as np
def distance(x, y, p_norm=2):
return np.sum(np.abs(x - y) ** p_norm) ** (1 / p_norm)
def sample_inside_sphere(dimensionality, radius, p_norm=2):
direction_unit_vector = (2 * np.random.rand(dimensionality) - 1)
direction_unit_vector = direction_unit_vector / distance(direction_unit_vector, np.zeros(dimensionality), p_norm)
return direction_unit_vector * np.random.rand() * radius
def rbf(d, gamma):
if gamma == 0.0:
return 0.0
else:
return np.exp(-(d / gamma) ** 2)
def rbf_score(point, minority_points, gamma, p_norm=2):
result = 0.0
for minority_point in minority_points:
result += rbf(distance(point, minority_point, p_norm), gamma)
return result
class RBCCR:
def __init__(self, energy, gamma=1.0, n_samples=100, threshold=0.33,
regions='E', p_norm=2, minority_class=None, n=None,
random_state=None, keep_appended=False, keep_radii=False):
self.energy = energy
self.gamma = gamma
self.n_samples = n_samples
self.threshold = threshold
self.regions = regions
self.p_norm = p_norm
self.minority_class = minority_class
self.n = n
self.random_state = random_state
self.keep_appended = keep_appended
self.keep_radii = keep_radii
self.appended = None
self.radii = None
def fit_sample(self, X, y):
np.random.seed(self.random_state)
if self.minority_class is None:
classes = np.unique(y)
sizes = [sum(y == c) for c in classes]
minority_class = classes[np.argmin(sizes)]
else:
minority_class = self.minority_class
minority_points = X[y == minority_class].copy()
majority_points = X[y != minority_class].copy()
minority_labels = y[y == minority_class].copy()
majority_labels = y[y != minority_class].copy()
if self.n is None:
n = len(majority_points) - len(minority_points)
else:
n = self.n
distances = np.zeros((len(minority_points), len(majority_points)))
for i in range(len(minority_points)):
for j in range(len(majority_points)):
distances[i][j] = distance(minority_points[i], majority_points[j], self.p_norm)
radii = np.zeros(len(minority_points))
translations = np.zeros(majority_points.shape)
for i in range(len(minority_points)):
minority_point = minority_points[i]
remaining_energy = self.energy
radius = 0.0
sorted_distances = np.argsort(distances[i])
n_majority_points_within_radius = 0
while True:
if n_majority_points_within_radius == len(majority_points):
if n_majority_points_within_radius == 0:
radius_change = remaining_energy / (n_majority_points_within_radius + 1)
else:
radius_change = remaining_energy / n_majority_points_within_radius
radius += radius_change
break
radius_change = remaining_energy / (n_majority_points_within_radius + 1)
if distances[i, sorted_distances[n_majority_points_within_radius]] >= radius + radius_change:
radius += radius_change
break
else:
if n_majority_points_within_radius == 0:
last_distance = 0.0
else:
last_distance = distances[i, sorted_distances[n_majority_points_within_radius - 1]]
radius_change = distances[i, sorted_distances[n_majority_points_within_radius]] - last_distance
radius += radius_change
remaining_energy -= radius_change * (n_majority_points_within_radius + 1)
n_majority_points_within_radius += 1
radii[i] = radius
for j in range(n_majority_points_within_radius):
majority_point = majority_points[sorted_distances[j]]
d = distances[i, sorted_distances[j]]
while d < 1e-20:
majority_point += (1e-6 * np.random.rand(len(majority_point)) + 1e-6) * \
np.random.choice([-1.0, 1.0], len(majority_point))
d = distance(minority_point, majority_point)
translation = (radius - d) / d * (majority_point - minority_point)
translations[sorted_distances[j]] += translation
if self.keep_radii:
self.radii = radii
appended = []
for i in range(len(minority_points)):
minority_point = minority_points[i]
n_synthetic_samples = int(np.round(1.0 / (radii[i] * np.sum(1.0 / radii)) * n))
r = radii[i]
if self.gamma is None or ('L' in self.regions and 'E' in self.regions and 'H' in self.regions):
for _ in range(n_synthetic_samples):
appended.append(minority_point + sample_inside_sphere(len(minority_point), r, self.p_norm))
else:
samples = []
scores = []
for _ in range(self.n_samples):
sample = minority_point + sample_inside_sphere(len(minority_point), r, self.p_norm)
score = rbf_score(sample, minority_points, self.gamma, self.p_norm)
samples.append(sample)
scores.append(score)
seed_score = rbf_score(minority_point, minority_points, self.gamma, self.p_norm)
lower_threshold = seed_score - self.threshold * (seed_score - np.min(scores + [seed_score]))
higher_threshold = seed_score + self.threshold * (np.max(scores + [seed_score]) - seed_score)
suitable_samples = [minority_point]
for sample, score in zip(samples, scores):
if score <= lower_threshold:
case = 'L'
elif score >= higher_threshold:
case = 'H'
else:
case = 'E'
if case in self.regions:
suitable_samples.append(sample)
suitable_samples = np.array(suitable_samples)
if n_synthetic_samples <= len(suitable_samples):
replace = False
else:
replace = True
selected_samples = suitable_samples[
np.random.choice(len(suitable_samples), n_synthetic_samples, replace=replace)
]
for sample in selected_samples:
appended.append(sample)
appended = np.array(appended)
if self.keep_appended:
self.appended = appended
majority_points += translations
if len(appended) > 0:
points = np.concatenate([majority_points, minority_points, appended])
labels = np.concatenate([majority_labels, minority_labels, np.tile([minority_class], len(appended))])
else:
points = np.concatenate([majority_points, minority_points])
labels = np.concatenate([majority_labels, minority_labels])
return points, labels