python-tools/stats.py at master · BarzinM/python-tools · 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
163
164
165
166
167
168
169
170
from collections import deque
import numpy as np


class RunningStats(object):

    def __init__(self, tau=None, *args):
        if len(args):
            print("WARNING: Remove extra argumetns when initializing RunningStates.")
        self.tau = tau
        self._func = self._first

    def __call__(self, *args):
        return self._func(*args)

    def _first(self, value=None, axis=0):
        self.avg = np.mean(value, axis=0)
        self.var = np.var(value, axis=0)
        self.data_count = len(value)

        if self.tau is None:
            self._func = self._exact
        else:
            self._func = self._running

        return self.avg, self.var

    def _exact(self, value=None, axis=0):
        if value is not None:
            n = len(value)

            new_data_mean = np.mean(value, axis=0)

            temp = value - new_data_mean
            new_data_var = np.dot(temp, temp) / n

            new_data_mean_sq = np.square(new_data_mean)

            new_means = ((self.avg * self.data_count) +
                         (new_data_mean * n)) / (self.data_count + n)

            self.var = (((self.data_count * (self.var + np.square(self.avg))) +
                         (n * (new_data_var + new_data_mean_sq))) / (self.data_count + n) -
                        np.square(new_means))

            # occasionally goes negative, clip
            self.var = np.maximum(0.0, self.var)
            self.avg = new_means
        return self.avg, self.var

    def _running(self, value=None, axis=0):
        if value is not None:
            var = np.var(value - self.avg, axis=0)
            self.var = self.var * (1 - self.tau) + var * self.tau
            self.avg = self.avg * (1 - self.tau) + value * self.tau
        return self.avg, self.var

    def normalize(self, values):
        return (values - self.avg) / (np.sqrt(self.var) + 1e-6)


class RunningAverage(RunningStats):

    def _first(self, value=None, axis=0):
        if type(value) in [int, float]:
            self.avg = value
            self.data_count = 1
        else:
            self.avg = np.mean(value, axis=0)
            self.data_count = len(value)

        if self.tau is None:
            self._func = self._exact
        else:
            self._func = self._running

        return self.avg

    def _exact(self, value=None, axis=0):
        if value is not None:
            n = len(value)

            new_data_mean = np.mean(value, axis=0)

            new_means = ((self.avg * self.data_count) +
                         (new_data_mean * n)) / (self.data_count + n)

            self.avg = new_means
        return self.avg

    def _running(self, value=None, axis=0):
        if value is not None:
            self.avg = self.avg * (1 - self.tau) + value * self.tau
        return self.avg


class Normality(object):

    def __init__(self, tol, window):
        self.window = window
        self.tol = tol
        self.var = 0.
        self.avg = 0.
        self.avg_var = 0.
        self.dist = [0, 0, 0, 0]
        self.queue = deque(maxlen=window)

    def update(self, value):
        total = len(self.queue)
        if total < self.window:
            var = np.linalg.norm(value - self.avg)
            self.avg = (self.avg * total + value) / \
                (total + 1)
            self.var = (self.var * total + var) / \
                (total + 1)
            z_score = abs(value - self.avg) / self.var

            if z_score < 1.:
                self.dist[0] += 1
                self.queue.append((0, value, var))
            elif z_score < 2.:
                self.dist[1] += 1
                self.queue.append((1, value, var))
            elif z_score < 3.:
                self.dist[2] += 1
                self.queue.append((2, value, var))
            else:
                self.dist[3] += 1
                self.queue.append((3, value, var))

            return False

        d, old_value, old_var = self.queue.popleft()
        self.dist[d] -= 1

        var = np.linalg.norm(value - self.avg)
        self.avg += (value - old_value) / total
        self.var += (var - old_var) / total

        z_score = abs(value - self.avg) / self.var

        if z_score < 1.:
            self.dist[0] += 1
            self.queue.append((0, value, var))
        elif z_score < 2.:
            self.dist[1] += 1
            self.queue.append((1, value, var))
        elif z_score < 3.:
            self.dist[2] += 1
            self.queue.append((2, value, var))
        else:
            self.dist[3] += 1
            self.queue.append((3, value, var))

        if abs(self.dist[0] / total - .6827) < self.tol and \
                abs(sum(self.dist[:2]) / total - .9545) < self.tol and \
                abs(sum(self.dist[:3]) / total - .9973) < self.tol:
            return True
        return False


def cosineSimilarity(a, b):
    return np.dot(a, b) / (np.norm(a) * np.norm(b))


if __name__ == "__main__":
    from time import time
    m = RunningStats()
    for _ in range(2):
        print(m([1, 2]))