Wasserstein-Learning-For-Point-Process/Plotter.py at master · xiaoshuai09/Wasserstein-Learning-For-Point-Process · 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
import numpy as np
import scipy.stats
import matplotlib
from _struct import Struct
matplotlib.use('agg')
import matplotlib.pyplot as plt

def read_sequences(seq_file):
    with open(seq_file) as f:
        lines = f.readlines()
    sequences = []
    for line in lines:
        x = line.split()
        seq = [None]*len(x)
        for i in range(len(x)):
            seq[i] = float(x[i])
        sequences.append(seq)
    f.close()
    return sequences

def  get_intensity(sequences, T = None, n_t=None, t0=None):
    if T is None:
        T = max(max(sequences))
    if n_t is None:
        n_t = 50
    if t0 is None:
        t0 = 0

    dt = (T-t0)/n_t
    ts = np.arange(t0,T,dt)
    n_seqs = len(sequences)
    lens = np.zeros((n_seqs,1))
    cnt = np.zeros((n_t,1))

    for i in range(n_seqs):
        seq = sequences[i]
        j = 0
        k = 0
        for t in np.arange(t0+dt,T+dt,dt):
            while (j < len(seq) and seq[j] <= t):
                j = j + 1
            cnt[k] = cnt[k] + j
            k = k + 1

            #print(t)
    dif = np.zeros((len(cnt),1))
    dif[0] = cnt[0]
    for i in range(len(cnt)-1):
        dif[i+1] = cnt[i+1]-cnt[i]
    intensity = dif/(n_seqs)/dt

    return ts, intensity


def get_integral_empirical(sequences, intensity, T, n_t, t0=None):

    if T is None:
        T = max(max(sequences))
    if n_t is None:
        n_t = 50
    if t0 is None:
        t0 = 0
    dt = (T-t0)/n_t
    ts = np.arange(t0,T,dt)
    n_seqs = len(sequences)


    integral = []
    for i in range(1000):
        seq = sequences[i]
        integral_seq = []
        for j in range(len(seq)-1):
            t_start = seq[j]
            t_end = seq[j+1]
            index_start = np.int( t_start/dt)
            index_end = np.int(t_end/dt)+1
            integral_seq.append( np.sum(intensity[index_start:index_end])*dt -intensity[index_start]*(t_start-index_start*dt)-intensity[index_end-1]*(index_end*dt-t_end))
            #-intensity[index_start]*(t_start-index_start*dt)-intensity[index_end-1]*(index_end*dt-t_end)

        integral += integral_seq
    return integral


def hawkes_integral(sequences,model):
    integrals = []
    for seq in sequences:
        integral = []
        seq = np.asarray(seq)
        for i in range(len(seq)-1):
            integral_delta = (seq[i+1]-seq[i])*model['mu'] + model['alpha'] * np.sum(np.exp(-(seq[i]-seq[:i+1]))-np.exp(-(seq[i+1]-seq[:i+1])))
            integral.append(integral_delta)
        integrals+=integral
    return integrals

def selfcorrecting_integral(sequences,model):
    integrals = []
    for seq in sequences:
        integral = []
        seq = np.asarray(seq)
        for i in range(len(seq)-1):
            integral_delta = (np.exp(model['mu']*seq[i+1]) - np.exp(model['mu']*seq[i]))/np.exp(model['alpha']*len(seq[:i+1]))/model['mu']
            integral.append(integral_delta)
        integrals+=integral
    return integrals

def gaussian_integral(sequences,model):
    integrals = []
    for seq in sequences:
        integral = []
        seq = np.asarray(seq)
        for i in range(len(seq)-1):
            integral_delta = np.sum( model['coef'] * (scipy.stats.norm.cdf(seq[i+1], model['center'], model['std']) - scipy.stats.norm.cdf(seq[i], model['center'], model['std']) ) )
            integral.append(integral_delta)
        integrals+=integral
    return integrals

def get_integral(sequences, data):
    sequences = sequences[:1000] # random sample?
    if data=='hawkes':
          model = dict()
          model['mu'] = 1
          model['w'] = 1
          model['alpha'] = 0.8
          integrals = hawkes_integral(sequences, model)
    elif data=='selfcorrecting':
          model = dict()
          model['mu'] = 1
          model['alpha'] = 0.2
          integrals = selfcorrecting_integral(sequences, model)
    elif data=='gaussian':
          model = dict()
          model['coef'] = [2,3,2]
          model['center'] = [3,7,11]
          model['std'] = [1,1,1]
          integrals = gaussian_integral(sequences, model)
    return integrals

if __name__ == '__main__':
    seq_file = 'real_data_sequences.txt'
    sequences = read_sequences(seq_file)

    seq_file_gen = 'real_data_sequences.txt'
    sequences_gen = read_sequences(seq_file_gen)

    T = 800
    n_t  = 30
    ts, intensity = get_intensity(sequences, T, n_t)
    ts_gen, intensity_gen = get_intensity(sequences_gen, T, n_t)

    plt.plot(ts,intensity, label='real')
    plt.plot(ts_gen, intensity_gen, label='generated')
    plt.legend(loc=1)
    plt.xlabel('time')
    plt.ylabel('intensity')