def __init__(self, coeff_list):
# NB: we adopt the weak-term-first convention for inputs
self.coeff_list = coeff_list
self.q = pylab.poly1d(coeff_list[::-1])
self.qd = pylab.polyder(self.q)
self.qdd = pylab.polyder(self.qd)
self.degree = self.q.order
def __init__(self, coeff_list):
# NB: we adopt the weak-term-first convention for inputs
self.coeff_list = coeff_list
self.q = pylab.poly1d(coeff_list[::-1])
self.qd = pylab.polyder(self.q)
self.qdd = pylab.polyder(self.qd)
self.degree = self.q.order
def do_one_regression(data, metadata):
"""do_one_regression
This determines the scatter of the fit at a single wavelength for all stars
Parameters
----------
data:
metadata:
returns
-------
"""
ln_s_values = np.arange(np.log(0.0001), 0., 0.5)
chis_eval = np.zeros_like(ln_s_values)
for ii, ln_s in enumerate(ln_s_values):
foo, bar, chi, logdet_Cinv = do_one_regression_at_fixed_scatter(data, metadata, scatter = np.exp(ln_s))
chis_eval[ii] = np.sum(chi * chi) - logdet_Cinv
if np.any(np.isnan(chis_eval)):
s_best = np.exp(ln_s_values[-1])
return do_one_regression_at_fixed_scatter(data, metadata, scatter = s_best) + (s_best, )
lowest = np.argmin(chis_eval)
if lowest == 0 or lowest == len(ln_s_values) + 1:
s_best = np.exp(ln_s_values[lowest])
return do_one_regression_at_fixed_scatter(data, metadata, scatter = s_best) + (s_best, )
ln_s_values_short = ln_s_values[np.array([lowest-1, lowest, lowest+1])]
chis_eval_short = chis_eval[np.array([lowest-1, lowest, lowest+1])]
z = np.polyfit(ln_s_values_short, chis_eval_short, 2)
f = np.poly1d(z)
fit_pder = np.polyder(z)
fit_pder2 = pylab.polyder(fit_pder)
s_best = np.exp(np.roots(fit_pder)[0])
return do_one_regression_at_fixed_scatter(data, metadata, scatter = s_best) + (s_best, )
def do_one_regression(data, metadata):
"""do_one_regression
This determines the scatter of the fit at a single wavelength for all stars
Parameters
----------
data:
metadata:
returns
-------
"""
ln_s_values = np.arange(np.log(0.0001), 0., 0.5)
chis_eval = np.zeros_like(ln_s_values)
for ii, ln_s in enumerate(ln_s_values):
foo, bar, chi, logdet_Cinv = do_one_regression_at_fixed_scatter(
data, metadata, scatter=np.exp(ln_s))
chis_eval[ii] = np.sum(chi * chi) - logdet_Cinv
if np.any(np.isnan(chis_eval)):
s_best = np.exp(ln_s_values[-1])
return do_one_regression_at_fixed_scatter(
data, metadata, scatter=s_best) + (s_best, )
lowest = np.argmin(chis_eval)
if lowest == 0 or lowest == len(ln_s_values) + 1:
s_best = np.exp(ln_s_values[lowest])
return do_one_regression_at_fixed_scatter(
data, metadata, scatter=s_best) + (s_best, )
ln_s_values_short = ln_s_values[np.array([lowest - 1, lowest, lowest + 1])]
chis_eval_short = chis_eval[np.array([lowest - 1, lowest, lowest + 1])]
z = np.polyfit(ln_s_values_short, chis_eval_short, 2)
f = np.poly1d(z)
fit_pder = np.polyder(z)
fit_pder2 = pylab.polyder(fit_pder)
s_best = np.exp(np.roots(fit_pder)[0])
return do_one_regression_at_fixed_scatter(data, metadata,
scatter=s_best) + (s_best, )
def Evaldn(self, s, n):
return pylab.polyder(self.q,n)(s)
