def test_biweight_location_constant_axis_3d():
shape = (10, 5, 2)
data = np.ones(shape)
cbl = biweight_location(data, axis=0)
assert_allclose(cbl, np.ones((shape[1], shape[2])))
cbl = biweight_location(data, axis=1)
assert_allclose(cbl, np.ones((shape[0], shape[2])))
cbl = biweight_location(data, axis=2)
assert_allclose(cbl, np.ones((shape[0], shape[1])))
def test_biweight_location_constant_axis_3d():
shape = (10, 5, 2)
data = np.ones(shape)
cbl = biweight_location(data, axis=0)
assert_allclose(cbl, np.ones((shape[1], shape[2])))
cbl = biweight_location(data, axis=1)
assert_allclose(cbl, np.ones((shape[0], shape[2])))
cbl = biweight_location(data, axis=2)
assert_allclose(cbl, np.ones((shape[0], shape[1])))
def test_biweight_location_ignore_nan():
data1d = np.array([1, 3, 5, 500, 2, np.nan])
data2d = np.array([data1d, data1d])
assert np.isnan(biweight_location(data1d, ignore_nan=False))
biw_expected = biweight_location(data1d[:-1], ignore_nan=False)
assert_equal(biweight_location(data1d, ignore_nan=True), biw_expected)
assert_equal(biweight_location(data2d, axis=0, ignore_nan=True), data1d)
assert_equal(biweight_location(data2d, axis=1, ignore_nan=True),
[biw_expected, biw_expected])
def test_biweight_location_constant_axis_2d():
shape = (10, 5)
data = np.ones(shape)
cbl = biweight_location(data, axis=0)
assert_allclose(cbl, np.ones(shape[1]))
cbl = biweight_location(data, axis=1)
assert_allclose(cbl, np.ones(shape[0]))
val1 = 100.
val2 = 2.
data = np.arange(50).reshape(10, 5)
data[2] = val1
data[7] = val2
cbl = biweight_location(data, axis=1)
assert_allclose(cbl[2], val1)
assert_allclose(cbl[7], val2)
def test_biweight_location_axis_3d():
"""Test a 3D array with the axis keyword."""
with NumpyRNGContext(12345):
nz = 3
ny = 4
nx = 5
data = np.random.normal(5, 2, (nz, ny, nx))
bw = biweight_location(data, axis=0)
assert bw.shape == (ny, nx)
y = 0
bwi = []
for i in range(nx):
bwi.append(biweight_location(data[:, y, i]))
bwi = np.array(bwi)
assert_allclose(bw[y], bwi)
def get_mean_cov(X, robust=True):
"""Get Mean Covariance.
Parameters
----------
X : array-like
robust : bool
Returns
-------
mean_tot : array-like
cov_tot : array-like
"""
from astropy.stats.biweight import (
biweight_midcovariance,
biweight_location,
)
if robust:
# Robust
mean_tot = biweight_location(X, axis=0)
cov_tot = biweight_midcovariance(X.T)
else:
# Classical
mean_tot = np.mean(X, axis=0)
cov_tot = np.cov(X)
return mean_tot, cov_tot
def test_biweight_location_constant_axis_2d():
shape = (10, 5)
data = np.ones(shape)
cbl = biweight_location(data, axis=0)
assert_allclose(cbl, np.ones(shape[1]))
cbl = biweight_location(data, axis=1)
assert_allclose(cbl, np.ones(shape[0]))
val1 = 100.
val2 = 2.
data = np.arange(50).reshape(10, 5)
data[2] = val1
data[7] = val2
cbl = biweight_location(data, axis=1)
assert_allclose(cbl[2], val1)
assert_allclose(cbl[7], val2)
def xbin_ybwt(x, y, xbins, Nmin=1):
"""
Take x,y pairs. Bin in x, find biweight location and scale
Input: x and y, xbins
Nmin : default 1
minimum number of points per bin to be used (otherwise nan)
Return: xbins centers, yloc, yscale
"""
assert len(x) == len(y)
xp = (xbins[1:]+xbins[:-1])/2
Nbins = len(xp)
yloc = np.zeros(Nbins) + np.nan
yscale = np.zeros(Nbins) + np.nan
bin_nums = np.digitize(x, xbins)
for ibin in range(Nbins):
bin_num = ibin + 1
vals = y[bin_nums == bin_num]
if len(vals) < Nmin: continue
yloc[ibin] = biweight_location(vals)
yscale[ibin] = biweight_scale(vals)
return xp, yloc, yscale
def test_biweight_location_axis_3d():
"""Test a 3D array with the axis keyword."""
with NumpyRNGContext(12345):
nz = 3
ny = 4
nx = 5
data = np.random.normal(5, 2, (nz, ny, nx))
bw = biweight_location(data, axis=0)
assert bw.shape == (ny, nx)
y = 0
bwi = []
for i in range(nx):
bwi.append(biweight_location(data[:, y, i]))
bwi = np.array(bwi)
assert_allclose(bw[y], bwi)
def test_biweight_location_nan():
data1d = np.array([1, 3, 5, 500, 2, np.nan])
all_nan = data1d.copy()
all_nan[:] = np.nan
data2d = np.array([data1d, data1d, all_nan])
data1d_masked = np.ma.masked_invalid(data1d)
data1d_masked.data[0] = np.nan
data2d_masked = np.ma.masked_invalid(data2d)
assert np.isnan(biweight_location(data1d))
bw_loc = biweight_location(data1d_masked)
assert not isinstance(bw_loc, np.ma.MaskedArray)
assert np.isnan(biweight_location(data2d))
for axis in (0, 1):
assert np.all(np.isnan(biweight_location(data2d, axis=axis)))
assert isinstance(biweight_location(data2d_masked, axis=axis),
np.ma.MaskedArray)
def test_biweight_location_axis():
"""Test a 2D array with the axis keyword."""
with NumpyRNGContext(12345):
ny = 100
nx = 200
data = np.random.normal(5, 2, (ny, nx))
bw = biweight_location(data, axis=0)
bwi = []
for i in range(nx):
bwi.append(biweight_location(data[:, i]))
bwi = np.array(bwi)
assert_allclose(bw, bwi)
bw = biweight_location(data, axis=1)
bwi = []
for i in range(ny):
bwi.append(biweight_location(data[i, :]))
bwi = np.array(bwi)
assert_allclose(bw, bwi)
def test_biweight_location_axis():
"""Test a 2D array with the axis keyword."""
with NumpyRNGContext(12345):
ny = 100
nx = 200
data = np.random.normal(5, 2, (ny, nx))
bw = biweight_location(data, axis=0)
bwi = []
for i in range(nx):
bwi.append(biweight_location(data[:, i]))
bwi = np.array(bwi)
assert_allclose(bw, bwi)
bw = biweight_location(data, axis=1)
bwi = []
for i in range(ny):
bwi.append(biweight_location(data[i, :]))
bwi = np.array(bwi)
assert_allclose(bw, bwi)
def filter_columns(data_2d):
# for each column, compute some robust average like the biweight
# print("data 2d shape ", data_2d.shape)
ny, nx = data_2d.shape
smdata = np.zeros(nx)
for i in range(nx):
coldata = data_2d[:, i].astype(float)
bwt = biweight.biweight_location(coldata)
smdata[i] = bwt
# print("data 1d shape ", smdata.shape)
return smdata
def find_resolution(multispec_fname,
initial_fwhm=.05,
usepercentile=True,
percentiles=[60, 80, 95],
Rguess=None,
full_output=False,
useclip=True,
findpeak=False,
makeplot=True):
"""
"""
from .spectrum import Spectrum1D
from astropy.stats import sigma_clip, biweight
from ..robust_polyfit import gaussfit
from ..utils import find_distribution_peak
import time
arcs = Spectrum1D.read(multispec_fname, flux_ext=4)
line_centers = np.loadtxt(os.path.dirname(__file__) +
"/../data/linelists/thar_list",
usecols=0)
start = time.time()
alllinefits = []
allRmed = []
allRerr = []
allwmid = []
allR = []
for i, arc in enumerate(arcs):
linefits = []
wave = arc.dispersion
flux = arc.flux
wmin, wmax = wave[0], wave[-1]
wmid = (wmin + wmax) / 2.
lcs = line_centers[(line_centers > wmin) & (line_centers < wmax)]
for lc in lcs:
fwhm = initial_fwhm
# get subpiece of arc
ii = (wave > lc - 5 * fwhm) & (wave < lc + 5 * fwhm)
_x, _y = wave[ii], flux[ii]
# guess amplitude, center, sigma
p0 = [np.max(_y), lc, fwhm / 2.355]
try:
popt = gaussfit(_x, _y, p0)
except:
pass
else:
if popt[0] > 0 and abs(popt[1] - lc) < .05:
linefits.append(popt)
try:
A, w, s = np.array(linefits).T
except ValueError:
print("This order did not have any good lines I guess")
#allR.append(np.nan); allRmed.append(np.nan); allRerr.append(np.nan); allwmid.append(wmid)
continue
alllinefits.append(linefits)
R = w / (s * 2.355)
if useclip: R = sigma_clip(R)
if findpeak:
if Rguess is None: Rguess = np.nanmedian(R)
try:
Rmed = find_distribution_peak(R, Rguess)
except (ValueError, RuntimeError):
print("--Could not find peak for arc {:02}".format(i))
print("--{}".format(sys.exc_info()))
Rmed = np.median(R)
elif usepercentile:
assert len(percentiles) == 3
Rlo, Rmed, Rhi = np.percentile(R, percentiles)
#Rerr = max(Rhi-Rmed, Rmed-Rlo)
Rerr = (Rmed - Rlo, Rhi - Rmed)
else:
Rmed = biweight.biweight_location(R)
Rerr = biweight.biweight_scale(R)
allR.append(R)
allRmed.append(Rmed)
allRerr.append(Rerr)
allwmid.append(wmid)
if usepercentile:
allRerr = np.array(allRerr).T
if makeplot:
import matplotlib.pyplot as plt
fig, ax = plt.subplots()
ax.errorbar(allwmid, allRmed, yerr=allRerr, fmt='o')
plt.show()
if full_output:
return allRmed, allRerr, allwmid, allR, arcs, alllinefits
return allRmed, allRerr, allwmid
def test_biweight_location_small():
cbl = biweight_location([1, 3, 5, 500, 2])
assert abs(cbl - 2.745) < 1e-3
def test_biweight_location_constant():
cbl = biweight_location(np.ones((10, 5)))
assert cbl == 1.
def test_biweight_location_small():
bw_loc = biweight_location([1, 3, 5, 500, 2])
assert_allclose(bw_loc, 2.7456117)
def test_biweight_location_masked():
data1d = np.array([1, 3, 5, 500, 2, np.nan])
data2d = np.array([data1d, data1d])
data1d_masked = np.ma.masked_invalid(data1d)
data2d_masked = np.ma.masked_invalid(data2d)
assert_equal(biweight_location(data1d, ignore_nan=True),
biweight_location(data1d_masked))
assert_equal(biweight_location(data2d, ignore_nan=True),
biweight_location(data2d_masked))
bw_loc = biweight_location(data2d, ignore_nan=True, axis=1)
bw_loc_masked = biweight_location(data2d_masked, axis=1)
assert isinstance(bw_loc_masked, np.ma.MaskedArray)
assert ~np.any(bw_loc_masked.mask) # mask is all False
assert_equal(bw_loc, bw_loc_masked.data)
bw_loc = biweight_location(data2d, ignore_nan=True, axis=0)
bw_loc_masked = biweight_location(data2d_masked, axis=0)
assert_equal(bw_loc_masked.data[:-1], bw_loc[:-1])
assert bw_loc_masked.mask[-1] # last mask element is True
data1d_masked.data[0] = np.nan # unmasked NaN
assert biweight_location(data1d_masked) is np.ma.masked
assert_equal(biweight_location(data1d_masked, ignore_nan=True),
biweight_location(data1d[1:], ignore_nan=True))
# ensure that input masked array is not modified
assert np.isnan(data1d_masked[0])
def test_biweight_location():
with NumpyRNGContext(12345):
# test that it runs
randvar = np.random.randn(10000)
cbl = biweight_location(randvar)
assert abs(cbl - 0) < 1e-2
def test_biweight_location_axis_tuple():
"""Test a 3D array with a tuple axis keyword."""
data = np.arange(24).reshape(2, 3, 4)
data[0, 0] = 100.
assert_equal(biweight_location(data, axis=0),
biweight_location(data, axis=(0,)))
assert_equal(biweight_location(data, axis=-1),
biweight_location(data, axis=(2,)))
assert_equal(biweight_location(data, axis=(0, 1)),
biweight_location(data, axis=(1, 0)))
assert_equal(biweight_location(data, axis=(0, 2)),
biweight_location(data, axis=(0, -1)))
assert_equal(biweight_location(data, axis=(0, 1, 2)),
biweight_location(data, axis=(2, 0, 1)))
assert_equal(biweight_location(data, axis=(0, 1, 2)),
biweight_location(data, axis=None))
def test_biweight_location():
with NumpyRNGContext(12345):
# test that it runs
randvar = np.random.randn(10000)
cbl = biweight_location(randvar)
assert abs(cbl - 0) < 1e-2
def test_biweight_location_constant():
cbl = biweight_location(np.ones((10, 5)))
assert cbl == 1.
def histogauss(SAMPLE,domain=[0,0]):
#+
#NAME:
# HISTOGAUSS
#
# PURPOSE:
# Histograms data and overlays it with a fitted Gaussian.
# Uses "MPFIT" to perform the fitting, which can be done on a restricted domain.
#
# CALLING SEQUENCE:
# model_params = histogauss.histogauss(mydata)
# model_params = histogauss.histogauss(mydata,[lower,upper])
#
# INPUT:
# mydata = nparray of data values (near-normally distributed) to be histogrammed
# domain = optional 2-valued list, specifying domain endpoints for Gaussian fit;
# default is to fit Gaussian to entire domain of data values
#
# OUTPUT ARGUMENTS:
# model_param = list of coefficients of the Gaussian fit:
#
# model_param[0]= the normalization ("height") of the fitted Gaussian
# model_param[1]= the mean of the fitted Gaussian
# model_param[2]= the standard deviation of the fitted Gaussian
# model_param[3]= the half-width of the 95% confidence interval
# of the biweight mean (not fitted)
#
# REVISION HISTORY:
# Written, H. Freudenreich, STX, 12/89
# More quantities returned in A, 2/94, HF
# Added NOPLOT keyword and print if Gaussian, 3/94
# Stopped printing confidence limits on normality 3/31/94 HF
# Added CHARSIZE keyword, changed annotation format, 8/94 HF
# Simplified calculation of Gaussian height, 5/95 HF
# Convert to V5.0, use T_CVF instead of STUDENT_T, GAUSSFIT instead of
# FITAGAUSS W. Landsman April 2002
# Correct call to T_CVF for calculation of A[3], 95% confidence interval
# P. Broos/W. Landsman July 2003
# Allow FONT keyword to be passed. T. Robishaw Apr. 2006
# Use Coyote Graphics for plotting W.L. Mar 2011
# Better formatting of text output W.L. May 2012
#-
# (Crudely) converted to Python3 by N. Grogin July 2020
#
DATA = SAMPLE.copy()
N = len(DATA)
# Make sure that not everything is in the same bin. If most
# data = 0, reject zeroes. If they are all some other value,
# complain and give up.
DATA.sort()
### boundaries of first and third quartiles
N3 = int(0.75*N)-1
N1 = int(0.25*N)-1
if (DATA[N3] == DATA[N1]):
if (DATA[int(N/2)-1] == 0):
if (sum(DATA!=0) > 15):
print('Suppressing Zeroes')
DATA = DATA[np.where(DATA!=0)]
N = len(DATA)
else:
print('Too Few Non-0 Values!')
return 0
else:
print('Too Many Identical Values: '+str(DATA[int(N/2)]))
return 0
# legacy structure from the IDL version; A[0] was an effective height
A = np.zeros(4)
NTOT = len(DATA)
# Outlier-resistant estimator of sample "mean":
A[1] = biweight_location(DATA)
# Outlier-resistant estimator of sample "standard deviation":
A[2] = biweight_scale(DATA)
# Compute he 95% confidence half-interval on the above mean:
M=0.7*(NTOT-1) #appropriate for a biweighted mean
CL = 0.95
two_tail_area = 1 - CL
A[3]=abs( sp_stats.t.ppf(1 - (two_tail_area)/2.0,M) ) * A[2] / np.sqrt(NTOT)
### clear the figure
plt.clf()
# Plot the histogram [important to have density=True for uniform normalization
# Also determines 'optimal' bin sizing, based on Freedman algorithm
histy,histx,ignored=hist(DATA, bins=200, histtype='stepfilled', density=True)
# Compute the midpoints of the histogram bins
xmid = np.zeros(len(histy))
for i in range(len(histy)):
xmid[i] = (histx[i]+histx[i+1])/2
# trim the histogram-fitting region, if a domain is specified
if (domain==[0,0]):
fitx = xmid
fity = histy
else:
fitx = xmid[(xmid>=domain[0]) & (xmid<=domain[1])]
fity = histy[(xmid>=domain[0]) & (xmid<=domain[1])]
# Array containing the initial guess of Gaussian params: normalization; mean; sigma
# !!! POOR RESULTS IF INPUT DATA ARE HIGHLY NON-GAUSSIAN !!!
p0 = [max(histy)*(A[2] * np.sqrt(2 * np.pi)),A[1],A[2]]
# Uniform weighting of each histogram bin value, in Gaussian fit
err = np.ones(len(fitx))
# prepare the dictionary of histogram information, for the fitting
fa = {'x':fitx, 'y':fity, 'err':err}
# perform the Gaussian fit to the histogram
m = mpfit.mpfit(gaussian_fitter, p0, functkw=fa)
# post-fitting diagnostics
print('mpfit status = ', m.status)
if (m.status <= 0):
print('mpfit error message = ', m.errmsg)
return 0
print('mpfit parameters = ', m.params)
[norm, mu, sigma] = m.params
### plot the model-fit Gaussian at finely-spaced locations spanning the data bins
### color the model-fitted region green, and the ignored region(s) red
finex = histx[0] + np.arange(1000) * (histx[-1]-histx[0])/1000
if (domain==[0,0]):
plt.plot(finex, norm/(sigma * np.sqrt(2 * np.pi)) *
np.exp( - (finex - mu)**2 / (2 * sigma**2) ),
linewidth=2, color='g')
else:
xsec = finex[(finex<domain[0])]
if len(xsec) > 0:
plt.plot(xsec, norm/(sigma * np.sqrt(2 * np.pi)) *
np.exp( - (xsec - mu)**2 / (2 * sigma**2) ),
linewidth=2, color='r')
xsec = finex[(finex>=domain[0]) & (finex<=domain[1])]
plt.plot(xsec, norm/(sigma * np.sqrt(2 * np.pi)) *
np.exp( - (xsec - mu)**2 / (2 * sigma**2) ),
linewidth=2, color='g')
xsec = finex[(finex>domain[1])]
if len(xsec) > 0:
plt.plot(xsec, norm/(sigma * np.sqrt(2 * np.pi)) *
np.exp( - (xsec - mu)**2 / (2 * sigma**2) ),
linewidth=2, color='r')
# NAG's devel environment does not easily allow plot-windowing,
# so the 'plt.show' is commented out, in favor of file-dumping
#plt.show()
### !!! BEWARE HARDCODED OUTPUT-FILENAME, BELOW !!!
### SUGGEST ADDING OUTPUT-FILENAME AS ADDITIONAL PARAMETER PASSED TO HISTOGAUSS
plt.savefig('temp.png')
return [norm, mu, sigma, A[3]]
def test_biweight_location_small():
cbl = biweight_location([1, 3, 5, 500, 2])
assert abs(cbl - 2.745) < 1e-3
