-
Notifications
You must be signed in to change notification settings - Fork 1
/
preliminary.py
165 lines (132 loc) · 6.31 KB
/
preliminary.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
import seaborn as sns
import numpy as np
import random
import os
import matplotlib.pyplot as plt
import mkl
import collections
from sklearn.neural_network import MLPClassifier
def main():
mkl.set_num_threads(4)
seed = 42
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
X_train, y_train = get_dataset1(300)
X_stream, y_stream = get_dataset1(3000)
plot_data(X_train, y_train, (2, 3, 1))
class_percentage = experiment(X_train, y_train, X_stream, y_stream)
plot_percentage(class_percentage, (2, 3, 2))
class_percentage = experiment_ballanced(X_train, y_train, X_stream, y_stream)
plot_percentage(class_percentage, (2, 3, 3))
X_train, y_train = get_dataset2(240, 60)
X_stream, y_stream = get_dataset2(2400, 600)
plot_data(X_train, y_train, (2, 3, 4))
class_percentage = experiment(X_train, y_train, X_stream, y_stream)
plot_percentage(class_percentage, (2, 3, 5))
class_percentage = experiment_ballanced(X_train, y_train, X_stream, y_stream)
plot_percentage(class_percentage, (2, 3, 6))
plt.show()
def get_dataset1(dataset_size):
size = dataset_size // 3
class1_data = np.random.multivariate_normal([-2, 1], [[5, 2], [2, 1]], size=size)
class2_data = np.random.multivariate_normal([2, 1], [[5, -2], [-2, 2]], size=size)
class3_data = np.random.multivariate_normal([0, -2.0], [[1.5, 0], [0, 1.5]], size=size)
X = np.concatenate((class1_data, class2_data, class3_data), axis=0)
y = np.concatenate((np.zeros(size), np.ones(size), np.ones(size)*2), axis=0)
idx = np.random.permutation(3*size).flatten()
X = X[idx]
y = y[idx]
return X, y
def get_dataset2(size1, size2):
class1_data = np.random.multivariate_normal([-1.5, 0], [[2, 1], [1, 2]], size=size1)
class2_data = np.random.multivariate_normal([1.5, 0], [[1, 0], [0, 1]], size=size2)
X = np.concatenate((class1_data, class2_data), axis=0)
y = np.concatenate((np.zeros(size1), np.ones(size2)), axis=0)
idx = np.random.permutation(size1+size2).flatten()
X = X[idx]
y = y[idx]
return X, y
def plot_data(X, y, subplot_idx):
with sns.axes_style("darkgrid"):
plt.subplot(*subplot_idx)
classes = np.unique(y)
colors = sns.color_palette("husl", 3) # ['#1f77b4', '#ff7f0e', '#2ca02c']
for c in classes:
x1, y1 = X[np.argwhere(y == c).flatten()].T
plt.plot(x1, y1, '.', color=colors[int(c)], label=f"class {int(c)+1}")
plt.legend()
def plot_percentage(class_percentage, subplot_idx):
with sns.axes_style("darkgrid"):
plt.subplot(*subplot_idx)
class_percentage = np.array(class_percentage)
cumulative = np.cumsum(class_percentage, axis=1)
num_classes = class_percentage.shape[1]
x = list(range(len(cumulative)))
colors = sns.color_palette("husl", 3) # ['#1f77b4', '#ff7f0e', '#2ca02c']
for i in range(num_classes-1, -1, -1):
plt.fill_between(x, np.zeros_like(x), cumulative[:, i], color=colors[i])
plt.xlabel('iterations')
plt.ylabel('class percentage')
plt.xlim(0, len(class_percentage))
plt.ylim(0, 1)
def experiment(X_train, y_train, X_stream, y_stream):
X_train_exp = np.copy(X_train)
y_train_exp = np.copy(y_train)
model = MLPClassifier(hidden_layer_sizes=(5, ), learning_rate_init=0.01, max_iter=100)
model.fit(X_train_exp, y_train_exp)
class_percentage = [[len(np.argwhere(y_train == i)) / len(y_train) for i in range(3)]]
for sample, target in zip(X_stream, y_stream):
sample = np.expand_dims(sample, axis=0)
target = np.expand_dims(target, axis=0)
y_pred = model.predict_proba(sample)
label = np.argmax(y_pred, axis=1)
if y_pred.max() < 0.7:
X_train_exp = np.concatenate((X_train_exp, sample), axis=0)
y_train_exp = np.concatenate((y_train_exp, target), axis=0)
model.partial_fit(X_train_exp, y_train_exp)
elif y_pred.max() > 0.95:
X_train_exp = np.concatenate((X_train_exp, sample), axis=0)
y_train_exp = np.concatenate((y_train_exp, label), axis=0)
model.partial_fit(X_train_exp, y_train_exp)
current_percentage = [len(np.argwhere(y_train_exp == i)) / len(y_train_exp) for i in range(3)]
class_percentage.append(current_percentage)
return class_percentage
def experiment_ballanced(X_train, y_train, X_stream, y_stream):
model = MLPClassifier(hidden_layer_sizes=(5, ), learning_rate_init=0.01, max_iter=100)
model.fit(X_train, y_train)
num_classes = len(np.unique(y_train).flatten())
last_predictions = collections.deque([], maxlen=50)
class_percentage = [[len(np.argwhere(y_train == i)) / len(y_train) for i in range(3)]]
for sample, target in zip(X_stream, y_stream):
sample = np.expand_dims(sample, axis=0)
target = np.expand_dims(target, axis=0)
y_pred = model.predict_proba(sample)
label = np.argmax(y_pred, axis=1)
if y_pred.max() < 0.7:
X_train = np.concatenate((X_train, sample), axis=0)
y_train = np.concatenate((y_train, target), axis=0)
last_predictions.append(int(target))
elif y_pred.max() > 0.9:
use_selflabeling = True
if len(last_predictions) >= 30:
current_dist = np.zeros(shape=(num_classes,), dtype=int)
class_label, class_count = np.unique(list(last_predictions), return_counts=True)
for i, count in zip(class_label, class_count):
current_dist[i] = count
current_dist = current_dist / len(last_predictions)
delta_p = current_dist[label] - (1.0 / num_classes)
if delta_p <= 0:
use_selflabeling = True
else:
use_selflabeling = False
if use_selflabeling:
last_predictions.append(int(label))
X_train = np.concatenate((X_train, sample), axis=0)
y_train = np.concatenate((y_train, label), axis=0)
model.partial_fit(X_train, y_train)
current_percentage = [len(np.argwhere(y_train == i)) / len(y_train) for i in range(3)]
class_percentage.append(current_percentage)
return class_percentage
if __name__ == '__main__':
main()