def test_fit_gaussians():
import numpy as np
import matplotlib.pyplot as plt
from myfitting import fit_gaussians
def gauss(x, amp, x0, width):
return amp * np.exp(-(x - x0)**2 / (2 * width**2))
x = np.linspace(-20, 15, 1000)
y = np.exp(-(x-3)**2 / (2 * 1**2)) + \
3 * np.exp(-(x-9)**2 / (2 * 0.5**2)) + \
5 * np.exp(-(x+0.5)**2 / (2 * 5**2))
# amp, offset, width
params = (1, 3, 1, 2, 9, 1, 0.5, -1, 1)
fit = fit_gaussians(x, y, ncomps=3, guesses=params)
assert len(fit) == 3
assert len(fit[1]) == 3
def calc_hi_vel_range(
hi_spectrum,
hi_vel_axis,
gauss_fit_kwargs,
width_scale=2,
co_spectrum=None,
co_vel_axis=None,
ncomps=1,
):
'''
Discussion of HI width error from Min:
(3) For the uncertainty in the HI velocity range, you take the absolute
difference between the Gaussian HI component center and the median CO peak
velocity. In fact, this is sort of the minimum uncertainty we expect. The
uncertainty mostly comes from our adoption of +/- 2 sigma(HI) to determine
the HI velocity range. In this sense, the maximum uncertainty we expect
would be (HI velocity range we determine by using +/- 2 sigma(HI) as
thresholds) - (HI velocity range over which CO emission appears).
Therefore, a more realistic estimate for the uncertainty in the HI velocity
range would be, e.g., ("minimum" error + "maximum" error) / 2. This
estimate would be a better measure in particular for Taurus, where CO
emission appears over a relatively small velocity range.
'''
from scipy.stats import nanmedian
from myfitting import fit_gaussians
co_width = np.copy(gauss_fit_kwargs['co_width'])
del (gauss_fit_kwargs['co_width'])
hi_fits = fit_gaussians(hi_vel_axis, hi_spectrum, **gauss_fit_kwargs)
gauss_fit_kwargs['co_width'] = co_width
# use either the gaussian closest to the CO peak, or the tallest gaussian
if co_spectrum is not None:
co_peak_vel = co_vel_axis[co_spectrum == np.nanmax(co_spectrum)][0]
vel_diffs_center = []
vel_diffs_edge = []
for i, param in enumerate(hi_fits[2]):
vel_center = param[1]
width = param[2]
vel_center_diff = np.abs(vel_center - co_peak_vel)
vel_diffs_center.append(vel_center_diff)
#vel_diffs_edge.append(vel_center_diff + 2.0 * width)
vel_diffs_edge.append(4.0 * width)
# get the velocity range
vel_diffs_center = np.asarray(vel_diffs_center)
vel_diffs_edge = np.asarray(vel_diffs_edge)
sort_indices = np.argsort(vel_diffs_center)
cloud_comp_num = np.asarray(sort_indices[:ncomps])
velocity_range = [np.inf, -np.inf]
for i in cloud_comp_num:
# get the component
param = hi_fits[2][i]
vel_center = param[1]
width = param[2]
# set absolute bounds if the component bounds extend beyond
upper_vel = vel_center + width * width_scale
lower_vel = vel_center - width * width_scale
if upper_vel > velocity_range[1]:
velocity_range[1] = upper_vel
if lower_vel < velocity_range[0]:
velocity_range[0] = lower_vel
# the offset between the co and fitted gaussians will be the HI error
hi_width_error_min = np.max(vel_diffs_center[cloud_comp_num])
# max error
hi_width_error_max = np.max(vel_diffs_edge[cloud_comp_num]) - \
co_width
#gauss_fit_kwargs['co_width']
hi_width_error_max = np.abs(velocity_range[1] - velocity_range[0]) - \
co_width
hi_width_error = (hi_width_error_min + hi_width_error_max) / 2.0
#print 'min error', hi_width_error_min, 'km/s'
#print 'max error', hi_width_error_max, 'km/s'
#print 'avg error', hi_width_error, 'km/s'
else:
amp_max = -np.Inf
for i, param in enumerate(hi_fits[2]):
if param[0] > amp_max:
amp_max = param[0]
vel_center = param[1]
width = param[2] * 4
cloud_comp_num = i
velocity_range = [
vel_center - width * width_scale, vel_center + width * width_scale
]
return velocity_range, hi_fits, cloud_comp_num, hi_width_error
def calc_hi_vel_range(hi_spectrum, hi_vel_axis, gauss_fit_kwargs,
width_scale=2, co_spectrum=None, co_vel_axis=None, ncomps=1,):
'''
Discussion of HI width error from Min:
(3) For the uncertainty in the HI velocity range, you take the absolute
difference between the Gaussian HI component center and the median CO peak
velocity. In fact, this is sort of the minimum uncertainty we expect. The
uncertainty mostly comes from our adoption of +/- 2 sigma(HI) to determine
the HI velocity range. In this sense, the maximum uncertainty we expect
would be (HI velocity range we determine by using +/- 2 sigma(HI) as
thresholds) - (HI velocity range over which CO emission appears).
Therefore, a more realistic estimate for the uncertainty in the HI velocity
range would be, e.g., ("minimum" error + "maximum" error) / 2. This
estimate would be a better measure in particular for Taurus, where CO
emission appears over a relatively small velocity range.
'''
from scipy.stats import nanmedian
from myfitting import fit_gaussians
co_width = np.copy(gauss_fit_kwargs['co_width'])
del(gauss_fit_kwargs['co_width'])
hi_fits = fit_gaussians(hi_vel_axis,
hi_spectrum, **gauss_fit_kwargs)
gauss_fit_kwargs['co_width'] = co_width
# use either the gaussian closest to the CO peak, or the tallest gaussian
if co_spectrum is not None:
co_peak_vel = co_vel_axis[co_spectrum == np.nanmax(co_spectrum)][0]
vel_diffs_center = []
vel_diffs_edge = []
for i, param in enumerate(hi_fits[2]):
vel_center = param[1]
width = param[2]
vel_center_diff = np.abs(vel_center - co_peak_vel)
vel_diffs_center.append(vel_center_diff)
#vel_diffs_edge.append(vel_center_diff + 2.0 * width)
vel_diffs_edge.append(4.0 * width)
# get the velocity range
vel_diffs_center = np.asarray(vel_diffs_center)
vel_diffs_edge = np.asarray(vel_diffs_edge)
sort_indices = np.argsort(vel_diffs_center)
cloud_comp_num = np.asarray(sort_indices[:ncomps])
velocity_range = [np.inf, -np.inf]
for i in cloud_comp_num:
# get the component
param = hi_fits[2][i]
vel_center = param[1]
width = param[2]
# set absolute bounds if the component bounds extend beyond
upper_vel = vel_center + width * width_scale
lower_vel = vel_center - width * width_scale
if upper_vel > velocity_range[1]:
velocity_range[1] = upper_vel
if lower_vel < velocity_range[0]:
velocity_range[0] = lower_vel
# the offset between the co and fitted gaussians will be the HI error
hi_width_error_min = np.max(vel_diffs_center[cloud_comp_num])
# max error
hi_width_error_max = np.max(vel_diffs_edge[cloud_comp_num]) - \
co_width
#gauss_fit_kwargs['co_width']
hi_width_error_max = np.abs(velocity_range[1] - velocity_range[0]) - \
co_width
hi_width_error = (hi_width_error_min + hi_width_error_max) / 2.0
#print 'min error', hi_width_error_min, 'km/s'
#print 'max error', hi_width_error_max, 'km/s'
#print 'avg error', hi_width_error, 'km/s'
else:
amp_max = -np.Inf
for i, param in enumerate(hi_fits[2]):
if param[0] > amp_max:
amp_max = param[0]
vel_center = param[1]
width = param[2] * 4
cloud_comp_num = i
velocity_range = [vel_center - width * width_scale,
vel_center + width * width_scale]
return velocity_range, hi_fits, cloud_comp_num, hi_width_error