-
Notifications
You must be signed in to change notification settings - Fork 0
/
simulate_signal.py
127 lines (96 loc) · 4.21 KB
/
simulate_signal.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
import math
from datetime import datetime
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
import torch
from torch.utils.data import DataLoader
from torch.utils.data import Dataset
import utils
from model.temporal_encoding import fourier_encoding, temporal_encoding
step_num = 1000
min_ts = 1419426000
max_ts = 1703422800
timestamps = np.linspace(min_ts, max_ts, num=step_num)
utils.global_seed(42)
sns.set_theme()
plt.rcParams['font.family'] = 'Times New Roman'
plt.rcParams['font.size'] = 14
plt.rcParams['figure.dpi'] = 300
def random_dynamics(x):
y = 2
result = []
for _ in x:
result.append(y)
y += np.random.normal(scale=0.05)
return np.array(result)
class SignalDataset(Dataset):
def __init__(self, timestamps, signal, d_hidden, min_ts, angular_velocity, is_fourier=False, pow_base=10000):
self.timestamp = timestamps
self.signal = signal
if is_fourier:
self.temporal_encodings = fourier_encoding(timestamps,
d_hidden,
min_ts,
angular_velocity)
else:
self.temporal_encodings = temporal_encoding(timestamps,
d_hidden,
min_ts,
angular_velocity,
pow_base)
def __len__(self):
return len(self.timestamp)
def __getitem__(self, idx):
return self.temporal_encodings[idx], self.signal[idx]
class TemporalSimulator(torch.nn.Module):
def __init__(self, d_hidden):
super(TemporalSimulator, self).__init__()
self.linear = torch.nn.Linear(d_hidden, 1)
def forward(self, x):
return self.linear(x)
def simulate_signal(timestamps, signal, min_ts, d_hidden, is_fourier=False, alpha=0.5, device='cuda', batch_num=50):
angular_velocity = 2 * math.pi * alpha / (max_ts - min_ts)
dataset = SignalDataset(timestamps, signal, d_hidden, min_ts, angular_velocity, is_fourier=is_fourier)
dataloader = DataLoader(dataset, batch_size=32)
model = TemporalSimulator(d_hidden).to(device)
criterion = torch.nn.MSELoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.98))
for i in range(batch_num):
loss_arr = []
for x, y_l in dataloader:
x = x.to(device)
y_l = y_l.to(device)
# print(y_l)
y_p = model.forward(x).squeeze(-1)
optimizer.zero_grad()
loss = criterion(y_p, y_l)
loss.backward()
optimizer.step()
loss_arr.append(loss.item())
print(f'loss: {sum(loss_arr) / len(loss_arr)}')
ts_tensor = torch.Tensor(dataset.temporal_encodings).to(device)
pred_signal = model.forward(ts_tensor).to('cpu').detach().numpy().reshape(signal.shape)
return pred_signal
def compare_simulate(timestamps, signal, min_ts, d_hidden, batch_num=50):
print("Fourier Series")
fs_signal = simulate_signal(timestamps, signal, min_ts, d_hidden, is_fourier=True, batch_num=batch_num)
print("Temporal Encodings")
te_signal = simulate_signal(timestamps, signal, min_ts, d_hidden, is_fourier=False, batch_num=batch_num)
plt.figure(figsize=(8, 4))
temporal_points = [datetime.fromtimestamp(ts) for ts in timestamps]
sns.lineplot(x=temporal_points, y=signal, label='Original Signal')
sns.lineplot(x=temporal_points, y=fs_signal, label='Fourier Series')
sns.lineplot(x=temporal_points, y=te_signal, label='Temporal Encodings')
plt.xlabel('Year')
plt.ylabel('Signal')
plt.legend()
plt.tight_layout()
plt.savefig(f'./image/simulate_{d_hidden}.png')
plt.show()
if __name__ == '__main__':
signal = np.array(random_dynamics(timestamps), dtype=np.float32)
plt.figure(figsize=(8, 4))
sns.lineplot(x=[datetime.fromtimestamp(ts) for ts in timestamps], y=signal, )
compare_simulate(timestamps, signal, min_ts, d_hidden=32, batch_num=100)
compare_simulate(timestamps, signal, min_ts, d_hidden=64, batch_num=100)