def scatter_res_raw(t, m, e, lomb_model):
""" From arXiv 1101_2406v1 Dubath 20110112 paper.
Scatter: res/raw
Median absolute deviation (MAD) of the residuals (obtained by subtracting
model values from the raw light curve) divided by the MAD of the raw
light-curve values around the median.
"""
lomb_resid = lomb_model["freq_fits"][-1]["resid"]
return cf.median_absolute_deviation(lomb_resid) / cf.median_absolute_deviation(m)
def p2p_model(x, y, frequency):
"""
Compute features that compare the residuals of data folded by estimated
period from Lomb-Scargle model with residuals folded by twice the estimated
period.
"""
sumsqr_diff_unfold = np.sum((np.diff(y)**2))
median_diff = np.median(np.abs(np.diff(y)))
mad = cf.median_absolute_deviation(y)
x = x.copy()
x = x - min(x)
t_2per_fold = np.array(x % (2. / frequency))
t_2per_sort_inds = np.argsort(t_2per_fold)
t_2per_fold = t_2per_fold[t_2per_sort_inds]
y_2per_fold = np.array(y)[t_2per_sort_inds]
sumsqr_diff_2per_fold = np.sum(np.diff(y_2per_fold)**2)
### eta feature from arXiv 1101.3316 Kim QSO paper:
t_1per_fold = np.array(x % (1. / frequency))
t_1per_sort_inds = np.argsort(t_1per_fold)
t_1per_fold = t_1per_fold[t_1per_sort_inds]
y_1per_fold = np.array(y)[t_1per_sort_inds]
median_1per_fold_diff = np.median(np.abs(np.diff(y_1per_fold)))
out_dict = {}
out_dict['scatter_2praw'] = sumsqr_diff_2per_fold / sumsqr_diff_unfold
out_dict['scatter_over_mad'] = median_diff / mad
out_dict['ssqr_diff_over_var'] = sumsqr_diff_unfold / ((len(y) - 1)
* np.var(y))
out_dict['scatter_pfold_over_mad'] = median_1per_fold_diff / mad
return out_dict
