def featvec(x_axis, sampledata, ticid=None, v=0):
"""calculates the feature vector of a single light curve
version 0: features 0-15
version 1: features 0-19
0 - Average - lower alpha
1 - Variance - upper case beta (B)
2 - Skewness - upper case gamma
3 - Kurtosis - upper case delta
4 - ln variance - lowercase beta
5 - ln skewness - lowercase gamma
6 - ln kurtosis - lowercase delta
(over 0.1 to 10 days)
7 - maximum power - upper episolon (E)
8 - ln maximum power - lower epsilon
9 - period of maximum power - upper zeta (Z)
10 - slope - upper eta (H)
11 - ln slope - lower eta
(integration of periodogram over time frame)
12 - P0 - 0.1-1 days - upper theta
13 - P1 - 1-3 days - upper iota
14 - P2 - 3-10 days - upper kappa
(over 0-0.1 days, for moving objects)
15 - Period of max power - upper lambda
(from transitleastsquares, OPTIONAL based on tls argument)
16 - best fit period - upper mu (M) - days
17 - best fit duration - lower mu - days
18 - best fit depth - upper nu (N) - ppt, measured from bottom
19 - power of best fit period - lower nu
for title purposes:
features_greek = [r'$\alpha$', 'B', r'$\Gamma$', r'$\Delta$', r'$\beta$', r'$\gamma$',r'$\delta$',
"E", r'$\epsilon$', "Z", "H", r'$\eta$', r'$\Theta$', "I", "K", r'$\Lambda$', "M", r'$\mu$'
,"N", r'$\nu$']
*** version 1 note: you may wish to go into the transitleastsquares's main.py file and
comment out all 'print' statements in order to save space while running this over lots of light curves
modified [lcg 07202020]"""
#empty feature vector
featvec = []
if v == 0:
#moments
featvec.append(
np.mean(sampledata)) #mean (don't use moment, always gives 0)
featvec.append(moment(sampledata, moment=2)) #variance
featvec.append(moment(sampledata, moment=3)) #skew
featvec.append(moment(sampledata, moment=4)) #kurtosis
featvec.append(np.log(np.abs(moment(sampledata,
moment=2)))) #ln variance
featvec.append(np.log(np.abs(moment(sampledata, moment=3)))) #ln skew
featvec.append(np.log(np.abs(moment(sampledata,
moment=4)))) #ln kurtosis
#periods
f = np.linspace(0.6, 62.8,
5000) #period range converted to frequencies
periods = np.linspace(0.1, 10, 5000) #0.1 to 10 day period
pg = signal.lombscargle(x_axis, sampledata, f, normalize=True)
rel_maxes = argrelextrema(pg, np.greater)
powers = []
indexes = []
for n in range(len(rel_maxes[0])):
index = rel_maxes[0][n]
indexes.append(index)
power_level_at_rel_max = pg[index]
powers.append(power_level_at_rel_max)
max_power = np.max(powers)
index_of_max_power = np.argmax(powers)
index_of_f_max = rel_maxes[0][index_of_max_power]
f_max_power = f[index_of_f_max]
period_max_power = 2 * np.pi / f_max_power
featvec.append(max_power)
featvec.append(np.log(np.abs(max_power)))
featvec.append(period_max_power)
slope = stats.linregress(x_axis, sampledata)[0]
featvec.append(slope)
featvec.append(np.log(np.abs(slope)))
#integrates the whole 0.1-10 day range
integrating1 = np.trapz(pg[457:5000],
periods[457:5000]) #0.1 days to 1 days
integrating2 = np.trapz(pg[121:457], periods[121:457]) #1-3 days
integrating3 = np.trapz(pg[0:121], periods[0:121]) #3-10 days
featvec.append(integrating1)
featvec.append(integrating2)
featvec.append(integrating3)
#for 0.001 to 1 day periods
f2 = np.linspace(62.8, 6283.2,
20) #period range converted to frequencies
p2 = np.linspace(0.001, 0.1, 20) #0.001 to 1 day periods
pg2 = signal.lombscargle(x_axis, sampledata, f2, normalize=True)
rel_maxes2 = argrelextrema(pg2, np.greater)
powers2 = []
indexes2 = []
for n in range(len(rel_maxes2[0])):
index2 = rel_maxes2[0][n]
indexes2.append(index2)
power_level_at_rel_max2 = pg2[index2]
powers2.append(power_level_at_rel_max2)
max_power2 = np.max(powers2)
index_of_max_power2 = np.argmax(powers2)
index_of_f_max2 = rel_maxes2[0][index_of_max_power2]
f_max_power2 = f2[index_of_f_max2]
period_max_power2 = 2 * np.pi / f_max_power2
featvec.append(period_max_power2)
#print("done")
#tls
elif v == 1:
from transitleastsquares import transitleastsquares, period_grid, catalog_info
model = transitleastsquares(x_axis, sampledata)
if type(ticid) != type(None):
dt = np.max(x_axis) - np.min(x_axis)
ab, mass, mass_min, mass_max, radius, radius_min, radius_max = catalog_info(
TIC_ID=int(ticid))
# >> find smallest period grid
rm_set = []
grid_lengths = [period_grid(1, 1, dt).shape[0]]
rm_set.append([1, 1])
if not np.isnan(radius):
grid_lengths.append(period_grid(radius, 1, dt).shape[0])
rm_set.append([radius, 1])
if not np.isnan(mass):
grid_lengths.append(period_grid(1, mass, dt).shape[0])
rm_set.append([1, mass])
ind = np.argmin(grid_lengths)
R_star, M_star = rm_set[ind]
else:
R_star, M_star = 1, 1
results = model.power(show_progress_bar=True,
R_star=R_star,
M_star=M_star)
featvec.append(results.period)
featvec.append(results.duration)
featvec.append((1 - results.depth))
featvec.append((results.power.max()))
return featvec
from __future__ import division, print_function
import numpy
from transitleastsquares import period_grid, duration_grid
if __name__ == "__main__":
print("Starting test: duration_grid...", end="")
periods = period_grid(
R_star=1, # R_sun
M_star=1, # M_sun
time_span=20, # days
period_min=0,
period_max=999,
oversampling_factor=3,
)
durations = duration_grid(periods, log_step=1.05, shortest=2)
numpy.testing.assert_almost_equal(max(durations), 0.12)
numpy.testing.assert_almost_equal(min(durations), 0.004562690993268325)
numpy.testing.assert_equal(len(durations), 69)
print("passed")
dirty_array[1] = None
dirty_array[2] = numpy.inf
dirty_array[3] = -numpy.inf
dirty_array[4] = numpy.nan
dirty_array[5] = -99
t, y, dy = cleaned_array(time_array, dirty_array, dy_array)
numpy.testing.assert_equal(len(t), 5)
numpy.testing.assert_equal(numpy.sum(t), 35)
numpy.testing.assert_equal(FAP(SDE=2), numpy.nan)
numpy.testing.assert_equal(FAP(SDE=7), 0.009443778)
numpy.testing.assert_equal(FAP(SDE=99), 8.0032e-05)
print("Test passed: FAP table")
periods = period_grid(R_star=1, M_star=1,
time_span=0.1) # R_sun # M_sun # days
numpy.testing.assert_almost_equal(max(periods), 2.4999999999999987)
numpy.testing.assert_almost_equal(min(periods), 0.6002621413799498)
numpy.testing.assert_equal(len(periods), 179)
periods = period_grid(R_star=1, M_star=1,
time_span=20) # R_sun # M_sun # days
numpy.testing.assert_almost_equal(max(periods), 10)
numpy.testing.assert_almost_equal(min(periods), 0.6009180713191087)
numpy.testing.assert_equal(len(periods), 1145)
periods = period_grid(
R_star=5, # R_sun
M_star=1, # M_sun
time_span=20, # days
period_min=0,