forked from YuliaRubanova/latent_ode
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgenerate_timeseries.py
140 lines (113 loc) · 4.15 KB
/
generate_timeseries.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
###########################
# Latent ODEs for Irregularly-Sampled Time Series
# Author: Yulia Rubanova
###########################
# Create a synthetic dataset
from __future__ import absolute_import, division
from __future__ import print_function
import os
import matplotlib
if os.path.exists("/Users/yulia"):
matplotlib.use('TkAgg')
else:
matplotlib.use('Agg')
import numpy as np
import numpy.random as npr
from scipy.special import expit as sigmoid
import pickle
import matplotlib.pyplot as plt
import matplotlib.image
import torch
import lib.utils as utils
# ======================================================================================
def get_next_val(init, t, tmin, tmax, final = None):
if final is None:
return init
val = init + (final - init) / (tmax - tmin) * t
return val
def generate_periodic(time_steps, init_freq, init_amplitude, starting_point,
final_freq = None, final_amplitude = None, phi_offset = 0.):
tmin = time_steps.min()
tmax = time_steps.max()
data = []
t_prev = time_steps[0]
phi = phi_offset
for t in time_steps:
dt = t - t_prev
amp = get_next_val(init_amplitude, t, tmin, tmax, final_amplitude)
freq = get_next_val(init_freq, t, tmin, tmax, final_freq)
phi = phi + 2 * np.pi * freq * dt # integrate to get phase
y = amp * np.sin(phi) + starting_point
t_prev = t
data.append([t,y])
return np.array(data)
def assign_value_or_sample(value, sampling_interval = [0.,1.]):
if value is None:
int_length = sampling_interval[1] - sampling_interval[0]
return np.random.random() * int_length + sampling_interval[0]
else:
return value
class TimeSeries:
def __init__(self, device = torch.device("cpu")):
self.device = device
self.z0 = None
def init_visualization(self):
self.fig = plt.figure(figsize=(10, 4), facecolor='white')
self.ax = self.fig.add_subplot(111, frameon=False)
plt.show(block=False)
def visualize(self, truth):
self.ax.plot(truth[:,0], truth[:,1])
def add_noise(self, traj_list, time_steps, noise_weight):
n_samples = traj_list.size(0)
# Add noise to all the points except the first point
n_tp = len(time_steps) - 1
noise = np.random.sample((n_samples, n_tp))
noise = torch.Tensor(noise).to(self.device)
traj_list_w_noise = traj_list.clone()
# Dimension [:,:,0] is a time dimension -- do not add noise to that
traj_list_w_noise[:,1:,0] += noise_weight * noise
return traj_list_w_noise
class Periodic_1d(TimeSeries):
def __init__(self, device = torch.device("cpu"),
init_freq = 0.3, init_amplitude = 1.,
final_amplitude = 10., final_freq = 1.,
z0 = 0.):
"""
If some of the parameters (init_freq, init_amplitude, final_amplitude, final_freq) is not provided, it is randomly sampled.
For now, all the time series share the time points and the starting point.
"""
super(Periodic_1d, self).__init__(device)
self.init_freq = init_freq
self.init_amplitude = init_amplitude
self.final_amplitude = final_amplitude
self.final_freq = final_freq
self.z0 = z0
def sample_traj(self, time_steps, n_samples = 1, noise_weight = 1.,
cut_out_section = None):
"""
Sample periodic functions.
"""
traj_list = []
for i in range(n_samples):
init_freq = assign_value_or_sample(self.init_freq, [0.4,0.8])
if self.final_freq is None:
final_freq = init_freq
else:
final_freq = assign_value_or_sample(self.final_freq, [0.4,0.8])
init_amplitude = assign_value_or_sample(self.init_amplitude, [0.,1.])
final_amplitude = assign_value_or_sample(self.final_amplitude, [0.,1.])
noisy_z0 = self.z0 + np.random.normal(loc=0., scale=0.1)
traj = generate_periodic(time_steps, init_freq = init_freq,
init_amplitude = init_amplitude, starting_point = noisy_z0,
final_amplitude = final_amplitude, final_freq = final_freq)
# Cut the time dimension
traj = np.expand_dims(traj[:,1:], 0)
traj_list.append(traj)
# shape: [n_samples, n_timesteps, 2]
# traj_list[:,:,0] -- time stamps
# traj_list[:,:,1] -- values at the time stamps
traj_list = np.array(traj_list)
traj_list = torch.Tensor().new_tensor(traj_list, device = self.device)
traj_list = traj_list.squeeze(1)
traj_list = self.add_noise(traj_list, time_steps, noise_weight)
return traj_list