|
1
|
|
|
# Licensed under a 3-clause BSD style license - see LICENSE |
|
|
|
|
|
|
2
|
|
|
"""Methods for synthetic generation of light curves.""" |
|
3
|
|
|
|
|
4
|
|
|
import logging |
|
5
|
|
|
|
|
6
|
|
|
import numpy as np |
|
|
|
|
|
|
7
|
|
|
import nfft |
|
|
|
|
|
|
8
|
|
|
import scipy.signal as scipy_signal |
|
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
from numba import jit |
|
|
|
|
|
|
11
|
|
|
|
|
12
|
|
|
__all__ = [ |
|
13
|
|
|
"lc_gen_samp", |
|
14
|
|
|
"lc_gen_psd_nft", |
|
15
|
|
|
"lc_gen_ou", |
|
16
|
|
|
"lc_gen_psd_lombscargle", |
|
17
|
|
|
"lc_gen_psd_fft", |
|
18
|
|
|
"lc_gen_psd_c", |
|
19
|
|
|
"fgen_wrapper", |
|
20
|
|
|
] |
|
21
|
|
|
|
|
22
|
|
|
log = logging.getLogger(__name__) |
|
23
|
|
|
|
|
24
|
|
|
|
|
25
|
|
|
def fgen_wrapper(fgen, t, y, fgen_params): |
|
|
|
|
|
|
26
|
|
|
"""Wrapper for all lc_gen_* functions, so as not to duplicate code.""" |
|
27
|
|
|
|
|
|
|
|
|
|
28
|
|
|
if fgen == "lc_gen_samp": |
|
29
|
|
|
y2 = lc_gen_samp(y) |
|
|
|
|
|
|
30
|
|
|
elif fgen == "lc_gen_psd_fft": |
|
31
|
|
|
y2 = lc_gen_psd_fft(y) |
|
|
|
|
|
|
32
|
|
|
elif fgen == "lc_gen_psd_nft": |
|
33
|
|
|
y2 = lc_gen_psd_nft(t, y) |
|
|
|
|
|
|
34
|
|
|
elif fgen == "lc_gen_psd_lombscargle": |
|
35
|
|
|
y2 = lc_gen_psd_lombscargle(t, y) |
|
|
|
|
|
|
36
|
|
|
elif fgen == "lc_gen_psd_c": |
|
37
|
|
|
y2 = lc_gen_psd_c(t, y, t) |
|
|
|
|
|
|
38
|
|
|
elif fgen == "lc_gen_ou": |
|
39
|
|
|
if (fgen_params is None) or not ('theta' in fgen_params or 'mu' in fgen_params or 'sigma' in fgen_params): |
|
|
|
|
|
|
40
|
|
|
raise Exception("You need to set the parameters for the signal") |
|
41
|
|
|
y2 = lc_gen_ou(times=t, **fgen_params) |
|
|
|
|
|
|
42
|
|
|
else: |
|
43
|
|
|
raise Exception(f"Unknown fgen method {fgen}") |
|
44
|
|
|
|
|
|
|
|
|
|
45
|
|
|
return y2 |
|
46
|
|
|
|
|
47
|
|
|
|
|
48
|
|
|
def lc_gen_samp(signs): |
|
49
|
|
|
"""Generation by sampling np.random.choice with same mean and std""" |
|
50
|
|
|
|
|
51
|
|
|
return np.random.choice(signs, signs.size) |
|
52
|
|
|
|
|
53
|
|
|
@jit(nopython=True) |
|
54
|
|
|
def lc_gen_ou(theta, mu, sigma, times, scale=None, loc=None): # pragma: no cover |
|
|
|
|
|
|
55
|
|
|
"""Generation from an OU process integrating the stochastic differential equation.""" |
|
56
|
|
|
|
|
57
|
|
|
width = 100 * times.size |
|
58
|
|
|
dt = (max(times) - min(times)) / width |
|
|
|
|
|
|
59
|
|
|
s2 = np.empty(times.size) |
|
|
|
|
|
|
60
|
|
|
s2[0] = mu # should get it from OU.rvs()!!!! |
|
61
|
|
|
for i in range(1, times.size): |
|
62
|
|
|
ti = times[i - 1] |
|
|
|
|
|
|
63
|
|
|
y = s2[i - 1] |
|
|
|
|
|
|
64
|
|
|
while ti < times[i]: |
|
65
|
|
|
y = y + dt * (theta * (mu - y) + sigma * y * np.random.randn() / np.sqrt(dt)) |
|
|
|
|
|
|
66
|
|
|
ti = ti + dt |
|
|
|
|
|
|
67
|
|
|
s2[i] = y |
|
68
|
|
|
if scale is not None: |
|
69
|
|
|
s2 = scale * s2 / np.std(s2) |
|
|
|
|
|
|
70
|
|
|
if loc is not None: |
|
71
|
|
|
s2 = s2 - np.mean(s2) + loc |
|
|
|
|
|
|
72
|
|
|
return s2 |
|
73
|
|
|
|
|
74
|
|
|
|
|
75
|
|
|
def lc_gen_psd_c(ts, values, times): |
|
|
|
|
|
|
76
|
|
|
"""Generation using interpolated PSD for light curves with similar PSD, mean and std.""" |
|
77
|
|
|
|
|
78
|
|
|
f, p = scipy_signal.welch(values, nperseg=ts.size / 2) |
|
|
|
|
|
|
79
|
|
|
fp = np.linspace(min(f), max(f), times.size // 2 + 1) |
|
|
|
|
|
|
80
|
|
|
pp = np.interp(fp, f, p) |
|
|
|
|
|
|
81
|
|
|
fft = np.sqrt(2 * pp * pp.size) * np.exp(1j * 2 * np.pi * np.random.random(pp.size)) |
|
82
|
|
|
s2 = np.fft.irfft(fft, n=values.size) |
|
|
|
|
|
|
83
|
|
|
a = values.std() / s2.std() |
|
|
|
|
|
|
84
|
|
|
b = values.mean() - a * s2.mean() |
|
|
|
|
|
|
85
|
|
|
s2 = a * s2 + b |
|
|
|
|
|
|
86
|
|
|
return s2 |
|
87
|
|
|
|
|
88
|
|
|
|
|
89
|
|
|
def lc_gen_psd_fft(values): |
|
90
|
|
|
"""Generation using Welch algorithm and the FFT, of synthetic signals with similar PSD, mean and std. |
|
|
|
|
|
|
91
|
|
|
|
|
92
|
|
|
Generates synthetic light curves using Lomb-Scargle algorithm |
|
93
|
|
|
to compute the power spectral density and the non-uniform fft |
|
94
|
|
|
to generate the signal.""" |
|
95
|
|
|
|
|
96
|
|
|
# this is not valid for non-uniform times (see PSD tests for a comparison) |
|
97
|
|
|
f, pxx = scipy_signal.welch(values) |
|
|
|
|
|
|
98
|
|
|
# fft2 = np.sqrt(2*Pxx*Pxx.size)*np.exp(1j*2*pi*np.random.randn(Pxx.size)) |
|
99
|
|
|
fft2 = np.sqrt(2 * pxx * pxx.size) * np.exp(1j * 2 * np.pi * np.random.random(pxx.size)) |
|
100
|
|
|
s2 = np.fft.irfft(fft2, n=values.size) |
|
|
|
|
|
|
101
|
|
|
a = values.std() / s2.std() |
|
|
|
|
|
|
102
|
|
|
b = values.mean() - a * s2.mean() |
|
|
|
|
|
|
103
|
|
|
s2 = a * s2 + b |
|
|
|
|
|
|
104
|
|
|
return s2 |
|
105
|
|
|
|
|
106
|
|
|
|
|
107
|
|
|
def lc_gen_psd_lombscargle(times, values): |
|
108
|
|
|
"""Generation using Lomb-Scargle algorithm and the non-uniform FFT, of synthetic signals with similar PSD, mean and std. |
|
|
|
|
|
|
109
|
|
|
|
|
110
|
|
|
Generates synthetic light curves using Lomb-Scargle algorithm |
|
111
|
|
|
to compute the power spectral density and the non-uniform fft |
|
112
|
|
|
to reconstruct the randomised signal.""" |
|
113
|
|
|
|
|
114
|
|
|
if values.size % 2 != 0: |
|
115
|
|
|
sigp = values[:-1] |
|
116
|
|
|
tp = times[:-1] |
|
|
|
|
|
|
117
|
|
|
else: |
|
118
|
|
|
sigp = values |
|
119
|
|
|
tp = times |
|
|
|
|
|
|
120
|
|
|
|
|
121
|
|
|
offset = 1e-6 |
|
122
|
|
|
n = sigp.size |
|
|
|
|
|
|
123
|
|
|
# k = np.arange(-n/2, n/2) no bc scipy_signal.lombscargle does not support freq zero |
|
124
|
|
|
k = np.linspace(-n / 2, n / 2 - 1 + offset, n) |
|
125
|
|
|
freqs = k / 2 / np.pi |
|
126
|
|
|
|
|
127
|
|
|
pxx = scipy_signal.lombscargle(tp, sigp, freqs) |
|
128
|
|
|
|
|
129
|
|
|
# build random phase to get real signal |
|
130
|
|
|
phase = np.random.random(pxx.size // 2) |
|
131
|
|
|
phase = np.concatenate((-np.flip(phase), [0], phase[:-1])) |
|
132
|
|
|
fft2 = np.sqrt(2 * pxx * pxx.size) * np.exp(1j * 2 * np.pi * phase) |
|
133
|
|
|
s2 = ( |
|
|
|
|
|
|
134
|
|
|
nfft.nfft((times - (times.max() + times.min()) / 2) / np.ptp(times), fft2, n, use_fft=True) |
|
135
|
|
|
/ n |
|
136
|
|
|
) |
|
137
|
|
|
|
|
138
|
|
|
# fix small deviations |
|
139
|
|
|
a = values.std() / s2.std() |
|
|
|
|
|
|
140
|
|
|
b = values.mean() - a * s2.mean() |
|
|
|
|
|
|
141
|
|
|
s2 = a * s2 + b |
|
|
|
|
|
|
142
|
|
|
|
|
143
|
|
|
s2 = np.real(s2) |
|
|
|
|
|
|
144
|
|
|
s2 = np.asarray(s2, dtype=float, order='C') # workaround for some bug in scipy lombscargle... |
|
|
|
|
|
|
145
|
|
|
|
|
|
|
|
|
|
146
|
|
|
return s2 |
|
147
|
|
|
|
|
148
|
|
|
def lc_gen_psd_nft(times, values): |
|
149
|
|
|
"""Generation using the non-uniform FFT of synthetic signals with similar PSD, mean and std. |
|
150
|
|
|
|
|
151
|
|
|
Generates synthetic light curves using the non-uniform FFT to |
|
152
|
|
|
compute the power spectral density and to reconstruct the |
|
153
|
|
|
randomised signal.""" |
|
154
|
|
|
|
|
155
|
|
|
k = np.arange(-times.size / 2, times.size / 2) |
|
156
|
|
|
n = 2* (k.size//2) |
|
|
|
|
|
|
157
|
|
|
|
|
158
|
|
|
nft = nfft.nfft_adjoint( |
|
159
|
|
|
(times - (times.max() + times.min()) / 2) / np.ptp(times), values, n, use_fft=True |
|
160
|
|
|
) |
|
161
|
|
|
|
|
162
|
|
|
# build random phase to get real signal |
|
163
|
|
|
phase = np.random.random(n // 2) |
|
164
|
|
|
phase = np.concatenate((-np.flip(phase), [0], phase[:-1])) |
|
165
|
|
|
|
|
166
|
|
|
fft2 = np.abs(nft) * np.exp(1j * 2 * np.pi * phase) |
|
167
|
|
|
s2 = ( |
|
|
|
|
|
|
168
|
|
|
nfft.nfft((times - (times.max() + times.min()) / 2) / np.ptp(times), fft2, use_fft=True) / n |
|
169
|
|
|
) |
|
170
|
|
|
s2 = np.real(s2) # np.real to fix small imaginary part from numerical error |
|
|
|
|
|
|
171
|
|
|
|
|
172
|
|
|
# fix small mean, std difference from numerical error |
|
173
|
|
|
a = values.std() / s2.std() |
|
|
|
|
|
|
174
|
|
|
b = values.mean() - a * s2.mean() |
|
|
|
|
|
|
175
|
|
|
s2 = a * s2 + b |
|
|
|
|
|
|
176
|
|
|
|
|
177
|
|
|
return s2 |
|
178
|
|
|
|
IAA-CSIC /
MUTIS
Loading history...
Cyclic imports may cause partly loaded modules to be returned. This might lead to unexpected runtime behavior which is hard to debug.