def calc_model(h_sd, alphaG):
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
return rh2_model
def odr_func(Z, h_sd):
return s14.calc_rh2(
h_sd,
alphaG,
Z,
phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
def calc_residual(params, h_sd, rh2):
alphaG = params['alphaG'].value
phi_g = params['phi_g'].value
Z = params['Z'].value
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False)
residual = rh2 - rh2_model
return residual
def odr_func(alphaG, h_sd,):
if 0:
plt.scatter(h_sd, rh2)
plt.yscale('log')
plt.ylim([1e-3, 1e2])
plt.annotate(alphaG, xy=(0.8, 0.1), xycoords='axes fraction')
plt.show()
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
return rh2_model
def calc_residual(params, h_sd, rh2, rh2_error=None):
alphaG = params['alphaG'].value
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
if rh2_error is not None:
residual = (rh2 - rh2_model) / rh2_error
else:
residual = rh2 - rh2_model
return residual
def calc_residual(params, h_sd, rh2):
alphaG = params['alphaG'].value
phi_g = params['phi_g'].value
Z = params['Z'].value
rh2_model = s14.calc_rh2(h_sd,
alphaG,
Z,
phi_g=phi_g,
return_fractions=False)
residual = rh2 - rh2_model
return residual
def calc_model(alphaG):
if 0:
plt.scatter(h_sd, rh2)
#plt.yscale('log')
#plt.ylim([1e-3, 1e2])
plt.annotate(alphaG, xy=(0.8, 0.1), xycoords='axes fraction')
plt.show()
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
#mask = rh2_model < 0
#print np.sum(rh2_model[~mask] < 0)
mask = np.zeros(np.size(rh2), dtype=bool)
chisq = np.nansum((rh2[~mask] - rh2_model[~mask])**2 / \
rh2_error[~mask]**2)
return chisq
def odr_func(Z, h_sd):
return s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
def fit_sternberg(h_sd, rh2, guesses=[10,], rh2_error=None,
h_sd_error=None, radiation_type='beamed', phi_g=2.0, Z=1.0,):
'''
Parameters
----------
h_sd : array-like
Hydrogen surface density in units of solar mass per parsec**2
rh2 : array-like
Ratio between molecular and atomic hydrogen masses.
guesses : None, scalar, or M-length sequence.
Initial guess for the parameters. See scipy.optimize.curve_fit.
rh2_error : bool
Error in rh2 parameter. Calculates a more accurate chi^2 statistic
Returns
-------
rh2_fit_params : array-like, optional
Model parameter fits.
Residual bootstrapping:
http://stats.stackexchange.com/questions/67519/bootstrapping-residuals-am-i-doing-it-right
'''
import sys
sys.path.insert(0, '/usr/users/ezbc/.local/lib64/python2.7/site-packages/')
from scipy.optimize import curve_fit, minimize
from scipy import stats
import scipy
#from lmfit import minimize, Parameters, report_fit, Minimizer
#import lmfit
from myscience import sternberg14 as s14
import mystats
import scipy.odr as odr
if any(np.isnan(h_sd)):
raise ValueError('HSD should not have nans')
mask = (rh2 < 0.0) | (np.isnan(h_sd))
mask = np.isnan(h_sd)
h_sd = h_sd[~mask]
rh2 = rh2[~mask]
rh2_error = rh2_error[~mask]
if 0:
def calc_residual(params, h_sd, rh2, rh2_error=None):
alphaG = params['alphaG'].value
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
if rh2_error is not None:
residual = (rh2 - rh2_model) / rh2_error
else:
residual = rh2 - rh2_model
return residual
# Set parameter limits and initial guesses
params = Parameters()
params.add('alphaG',
value=guesses[0],
min=0.01,
max=1000,
vary=True)
# Perform the fit!
result = minimize(calc_residual,
params,
args=(h_sd, rh2, rh2_error),
method='leastsq',
#method='anneal',
is_weighted=True,
)
# best-fit parameter
alphaG = params['alphaG'].value
elif 0:
def calc_model(h_sd, alphaG):
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
return rh2_model
popt, pcov = curve_fit(calc_model,
h_sd,
rh2,
p0=guesses,
sigma=rh2_error,
method='lm',
)
alphaG = popt[0]
elif 1:
def calc_model(alphaG):
if 0:
plt.scatter(h_sd, rh2)
#plt.yscale('log')
#plt.ylim([1e-3, 1e2])
plt.annotate(alphaG, xy=(0.8, 0.1), xycoords='axes fraction')
plt.show()
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
#mask = rh2_model < 0
#print np.sum(rh2_model[~mask] < 0)
mask = np.zeros(np.size(rh2), dtype=bool)
chisq = np.nansum((rh2[~mask] - rh2_model[~mask])**2 / \
rh2_error[~mask]**2)
return chisq
res = minimize(calc_model,
guesses[0],
#method='powell',
#method='nelder-mead',
method='L-BFGS-B',
bounds=[[0, None],],
options={'disp':False},
)
alphaG = res.x[0]
else:
def odr_func(alphaG, h_sd,):
if 0:
plt.scatter(h_sd, rh2)
plt.yscale('log')
plt.ylim([1e-3, 1e2])
plt.annotate(alphaG, xy=(0.8, 0.1), xycoords='axes fraction')
plt.show()
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
return rh2_model
model = odr.Model(odr_func)
data = odr.RealData(h_sd, rh2, sy=rh2_error)
odr_instance = odr.ODR(data, model, beta0=guesses)
output = odr_instance.run()
alphaG = output.beta[0]
# calculate model for observed H
rh2_model = s14.calc_rh2(h_sd, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
# calculate model for resampled H
h_sd_resampled = np.logspace(-3, 3, 1000)
rh2_resampled = s14.calc_rh2(h_sd_resampled, alphaG, Z, phi_g=phi_g,
return_fractions=False,
radiation_type=radiation_type,
)
return alphaG, rh2_model, rh2_resampled, h_sd_resampled
