SimpleAttention/dataset.py at main · donglinkang2021/SimpleAttention · GitHub

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
import numpy as np
from typing import Tuple

def regress_plane(
        num_samples:int,
        noise:float,
        radius:int = 6
    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Generates a dataset of points on a plane with a label
    corresponding to the distance from the origin.
    """
    def get_label(x, y):
        return (x + y) / (2 * radius)

    x = np.random.uniform(-radius, radius, num_samples)
    y = np.random.uniform(-radius, radius, num_samples)
    noise_x = np.random.uniform(-radius, radius, num_samples) * noise
    noise_y = np.random.uniform(-radius, radius, num_samples) * noise
    label = get_label(x + noise_x, y + noise_y)

    return x, y, label

def regress_gaussian(
        num_samples:int,
        noise:float,
        radius:int = 6
    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Generates a dataset of points with designed gaussian centers
    """
    def label_scale(distance):
        # distance less, label more
        # more probability of being closed to the center
        return distance * -0.1 + 1

    gaussians = np.array([
        [-4, 2.5, 1],
        [0, 2.5, -1],
        [4, 2.5, 1],
        [-4, -2.5, -1],
        [0, -2.5, 1],
        [4, -2.5, -1]
    ])
    gaussian_xy = gaussians[:, :2] # (num_gaussians, 2)
    gaussian_sign = gaussians[:, 2]

    x = np.random.uniform(-radius, radius, num_samples)
    y = np.random.uniform(-radius, radius, num_samples)
    noise_x = np.random.uniform(-radius, radius, num_samples) * noise
    noise_y = np.random.uniform(-radius, radius, num_samples) * noise
    tmp_xy = np.array([x + noise_x, y + noise_y]).T # (num_samples, 2)
    # we get the distance between our random points and the gaussian centers
    distance = np.sum((tmp_xy[:, np.newaxis] - gaussian_xy) ** 2, axis=-1) # (num_samples, num_gaussians)
    # select the gaussian center idx with the smallest distance
    idx = distance.argmin(axis=-1) # (num_samples,)
    # get the gaussian sign label for each sample
    label = (label_scale(distance) * gaussian_sign)[np.arange(num_samples), idx]

    return x, y, label

def classify_two_gauss_data(
        num_samples:int,
        noise:float,
        radius:int = 6
    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Generates a dataset of points with two gaussian centers
    """
    variance = radius * noise + 0.5
    n = num_samples // 2

    def gen_gauss(cx, cy, label):
        x = cx + np.sqrt(variance) * np.random.randn(n)
        y = cy + np.sqrt(variance) * np.random.randn(n)
        label = np.ones(n) * label
        return x, y, label

    x_pos, y_pos, label_pos = gen_gauss(2, 2, 1)
    x_neg, y_neg, label_neg = gen_gauss(-2, -2, -1)

    x = np.concatenate([x_pos, x_neg])
    y = np.concatenate([y_pos, y_neg])
    label = np.concatenate([label_pos, label_neg])

    return x, y, label

def classify_spiral_data(
        num_samples:int,
        noise:float,
        radius:int = 6
    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Generates a dataset of points with two spiral data
    """
    n = num_samples // 2

    def gen_spiral(delta_t, label):
        r = np.linspace(0.0, radius, n)
        t = 1.75 * np.linspace(0.0, 2 * np.pi, n) + delta_t
        x = r * np.sin(t) + np.random.uniform(-1, 1, n) * noise
        y = r * np.cos(t) + np.random.uniform(-1, 1, n) * noise
        label = np.ones(n) * label
        return x, y, label

    x_pos, y_pos, label_pos = gen_spiral(0, 1)
    x_neg, y_neg, label_neg = gen_spiral(np.pi, -1)
    x = np.concatenate([x_pos, x_neg])
    y = np.concatenate([y_pos, y_neg])
    label = np.concatenate([label_pos, label_neg])

    return x, y, label

def classify_circle_data(
        num_samples:int,
        noise:float,
        radius:int = 6
    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Generates a dataset of points with two circle data
    """
    n = num_samples // 2

    def gen_circle(min_r, max_r):
        r = np.random.uniform(min_r, max_r, n)
        t = np.random.uniform(0, 2 * np.pi, n)
        x = r * np.sin(t)
        y = r * np.cos(t)
        noise_x = np.random.uniform(-radius, radius, n) * noise
        noise_y = np.random.uniform(-radius, radius, n) * noise
        tmp_xy = np.array([x + noise_x, y + noise_y]).T # (num_samples, 2)
        distance = np.sqrt(np.sum(tmp_xy ** 2, axis=-1))
        label = np.ones(n) * (distance < (radius * 0.5)) * 2 - 1
        return x, y, label

    x_pos, y_pos, label_pos = gen_circle(0, radius * 0.5)
    x_neg, y_neg, label_neg = gen_circle(radius * 0.7, radius)
    x = np.concatenate([x_pos, x_neg])
    y = np.concatenate([y_pos, y_neg])
    label = np.concatenate([label_pos, label_neg])

    return x, y, label

def classify_xor_data(
        num_samples:int,
        noise:float,
        radius:int = 6,
        padding:float = 0.3
    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Generates a dataset of points with two xor data
    """

    x = np.random.uniform(-radius, radius, num_samples)
    y = np.random.uniform(-radius, radius, num_samples)
    x += ((x > 0) * 2 - 1) * padding
    y += ((y > 0) * 2 - 1) * padding
    noise_x = np.random.uniform(-radius, radius, num_samples) * noise
    noise_y = np.random.uniform(-radius, radius, num_samples) * noise
    label = ((x + noise_x) * (y + noise_y) >= 0) * 2 - 1

    return x, y, label