def fit_data_bg(x, y, peak_pos, peak_type='LO', width=None, bg_ord=0):
"""
Builds a lmfit model of peaks in listed by index in `peak_pos`
Parameters
----------
peak_type : string (default='lorentizian')
Peaks can be of the following types:
- 'LO' : symmetric lorentzian
- 'GA' : symmetric gaussain
- 'VO' : symmetric pseudo voigt
max_width : int (default = total points/10)
max width (in data points) that peak fitted can be
bg_ord: int
order of the background polynomial
0: constant, 1: linear, ...
Returns
-------
out:
fitted model
"""
# need to define peak width finding
if width is None:
width = guess_peak_width(x, y)
# start with polynomial background
model = PolynomialModel(bg_ord, prefix='bg_')
pars = model.make_params()
if peak_type == 'LO':
peak_function = lorentzian
elif peak_type == 'GA':
peak_function = gaussian
elif peak_type == 'VO':
peak_function = voigt
# add peak type for all peaks
for i, peak in enumerate(peak_pos):
temp_model = Model(peak_function, prefix='p%s_' % i)
pars.update(temp_model.make_params())
model += temp_model
# set initial background as flat line at zeros
for i in range(bg_ord + 1):
pars['bg_c%i' % i].set(0)
# give values for other peaks, keeping width and height positive
for i, peak in enumerate(peak_pos):
pars['p%s_x0' % i].set(x[peak])
pars['p%s_fwhm' % i].set(width, min=0)
pars['p%s_amp' % i].set(y[peak], min=0)
out = model.fit(y, pars, x=x)
return out
def fit_experimental_data_gauss(exp_x,
exp_y,
expected_peak_pos,
deg_of_bck_poly=5,
maxfev=25000):
num_of_peaks = len(expected_peak_pos)
mod = PolynomialModel(deg_of_bck_poly, prefix='poly_')
for c in range(num_of_peaks):
mod = mod + PseudoVoigtModel(prefix='p{}_'.format(c))
params = mod.make_params()
center = 0
sigma = 0
amplitude = 0
fraction = 0
for param in params:
if 'center' in param:
params[param].set(value=expected_peak_pos[center])
params[param].set(min=expected_peak_pos[center] - 0.5)
params[param].set(max=expected_peak_pos[center] + 0.5)
center += 1
if 'poly' in param:
if param == 'poly_c0':
params[param].set(value=50)
params[param].set(min=-100)
params[param].set(max=100)
continue
if param == 'poly_c1':
params[param].set(value=-1)
params[param].set(min=-100)
params[param].set(max=100)
continue
params[param].set(value=0)
## params[param].set(min = 3e-1)
## params[param].set(max = 3e-1)
if 'sigma' in param:
params[param].set(value=0.5)
params[param].set(min=0.0001)
params[param].set(max=0.8)
sigma += 1
if 'amplitude' in param:
params[param].set(value=5.5)
params[param].set(min=0.0001)
amplitude += 1
if 'fraction' in param:
params[param].set(value=0.0)
params[param].set(min=0.000)
params[param].set(max=0.000001)
fraction += 1
result = mod.fit(np.asarray(exp_y),
params,
x=np.asarray(exp_x),
fit_kws={'maxfev': maxfev})
print(result.fit_report())
return result
def fit_data_bg(x, y, peak_pos, peak_type="LO", width=None, bg_ord=0):
"""
Builds a lmfit model of peaks in listed by index in `peak_pos`
Parameters
----------
peak_type : string (default='lorentizian')
Peaks can be of the following types:
- 'LO' : symmetric lorentzian
- 'GA' : symmetric gaussain
- 'VO' : symmetric pseudo voigt
max_width : int (default = total points/10)
max width (in data points) that peak fitted can be
bg_ord: int
order of the background polynomial
0: constant, 1: linear, ...
Returns
-------
out:
fitted model
"""
# need to define peak width finding
if width is None:
width = guess_peak_width(x, y)
# start with polynomial background
model = PolynomialModel(bg_ord, prefix="bg_")
pars = model.make_params()
if peak_type == "LO":
peak_function = lorentzian
elif peak_type == "GA":
peak_function = gaussian
elif peak_type == "VO":
peak_function = voigt
# add peak type for all peaks
for i, peak in enumerate(peak_pos):
temp_model = Model(peak_function, prefix="p%s_" % i)
pars.update(temp_model.make_params())
model += temp_model
# set initial background as flat line at zeros
for i in range(bg_ord + 1):
pars["bg_c%i" % i].set(0)
# give values for other peaks, keeping width and height positive
for i, peak in enumerate(peak_pos):
pars["p%s_x0" % i].set(x[peak])
pars["p%s_fwhm" % i].set(width, min=0)
pars["p%s_amp" % i].set(y[peak], min=0)
out = model.fit(y, pars, x=x)
return out
def bg_sub(raman_spectra,
plot_each=False,
reduce_region=[3090, 4000],
cut_region=[3150, 3722],
order=2,
to_run='all'):
bg = PolynomialModel(order)
flag = False
if type(raman_spectra) == AHR.RamanSpectrum:
raman_spectra = {'a': raman_spectra}
flag = True
BGsub = {}
for key, spec in raman_spectra.items():
if to_run != 'all':
if key not in to_run:
continue
spec_red = spec.reduce_wn_region(reduce_region)
spec_cut = spec_red.cut_wn_region(cut_region)
spec_cut_x = spec_cut.wn
spec_cut_y = np.squeeze(spec_cut.spec_data.T)
bg_params = bg.make_params(c0=0,
c1=0,
c2=0,
c3=0,
c4=0,
c5=5,
c6=0,
c7=0)
bg_fit = bg.fit(spec_cut_y, x=spec_cut_x, params=bg_params)
spec_red_bg_sub = AHR.RamanSpectrum(
spec_red.wn,
np.squeeze(spec_red.spec_data.T) - bg_fit.eval(x=spec_red.wn))
BGsub[key] = spec_red_bg_sub
if plot_each:
fig, ax = plt.subplots()
ax.plot(spec_red.wn, spec_red.spec_data.T)
ax.scatter(spec_cut.wn, spec_cut.spec_data.T, s=3, color='orange')
ax.plot(spec_red.wn, bg_fit.eval(x=spec_red.wn))
ax.set_title(key)
if flag:
return BGsub['a']
else:
return BGsub
def prepare_for_fitting(self, poly_order, maxwidth, centerrange):
"""
:param x_center: numpy array of initial x values at picked centers
:param y_center: numpy array of initial y values at picked centers
:param fwhm: single float number for initial fwhm value
"""
self.set_baseline(poly_order)
baseline_mod = PolynomialModel(poly_order, prefix='b_')
mod = baseline_mod
pars = baseline_mod.make_params()
peakinfo = {}
for i in range(poly_order + 1):
prefix = "b_c{0:d}".format(i)
pars[prefix].set(value=self.baseline_in_queue[i]['value'],
vary=self.baseline_in_queue[i]['vary'])
i = 0
for peak in self.peaks_in_queue:
prefix = "p{0:d}_".format(i)
peak_mod = PseudoVoigtModel(prefix=prefix, )
pars.update(peak_mod.make_params())
pars[prefix + 'center'].set(value=peak['center'],
min=peak['center'] - centerrange,
max=peak['center'] + centerrange,
vary=peak['center_vary'])
pars[prefix + 'sigma'].set(value=peak['sigma'],
min=0.0,
vary=peak['sigma_vary'],
max=maxwidth)
pars[prefix + 'amplitude'].set(value=peak['amplitude'],
min=0,
vary=peak['amplitude_vary'])
pars[prefix + 'fraction'].set(value=peak['fraction'],
min=0.,
max=1.,
vary=peak['fraction_vary'])
peakinfo[prefix + 'phasename'] = peak['phasename']
peakinfo[prefix + 'h'] = peak['h']
peakinfo[prefix + 'k'] = peak['k']
peakinfo[prefix + 'l'] = peak['l']
mod += peak_mod
i += 1
self.parameters = pars
self.peakinfo = peakinfo
self.fit_model = mod
def fitTwoGaussians(x,y):
background = PolynomialModel(2)
pars = background.make_params()
peak1 = GaussianModel(prefix='p1_')
pars.update( peak1.make_params())
peak2 = GaussianModel(prefix='p2_')
pars.update( peak2.make_params())
# Guess some parameters from data to help the fitting
span = max(x)-min(x)
c1Guess = (y[-1]-y[0])/(x[-1]-x[0])
c0Guess = y[0]-c1Guess*x[0]
bgGuess = background.func(x=x,c0=c0Guess,c1=c1Guess,c2=0.)
signalGuess=min(y-bgGuess)
sigmaGuess = span/30.
amplitudeGuess = signalGuess*(sigmaGuess*np.sqrt(2.0*np.pi))
# Fit variables initialization
# pars.add('splitting',0.0001,max=span)
pars['c2'].set(0.,min=-0.000001,max=0.001)
pars['c1'].set(c1Guess)
pars['c0'].set(c0Guess)
pars['p1_center'].set(min(x)+span*0.35,min=min(x),max=max(x))
pars['p2_center'].set(min(x)+span*0.55,min=min(x),max=max(x))
# pars['p2_center'].set(min(x)+span*0.65,expr='p1_center+splitting')
pars['p1_amplitude'].set(amplitudeGuess,max=amplitudeGuess/10000.)
pars['p2_amplitude'].set(amplitudeGuess,max=amplitudeGuess/10000.)
pars['p1_sigma'].set(sigmaGuess, min=sigmaGuess/100.,max=sigmaGuess*10000.)
pars['p2_sigma'].set(sigmaGuess, min=sigmaGuess/100.,max=sigmaGuess*10000.)
#Add some useful parameters to evaluate
pars.add('p1_signal', expr='p1_amplitude/(p1_sigma*sqrt(2.0*pi))')
pars.add('p2_signal', expr='p2_amplitude/(p2_sigma*sqrt(2.0*pi))')
pars.add('p1_contrast', expr='-p1_amplitude/(p1_sigma*sqrt(2.0*pi)*(c0+c1*p1_center+c2*p1_center**2))')
pars.add('p2_contrast', expr='-p2_amplitude/(p2_sigma*sqrt(2.0*pi)*(c0+c1*p2_center+c2*p2_center**2))')
pars.add('splitting',pars['p2_center']-pars['p1_center'],expr='p2_center-p1_center',min=0.00001)
model = peak1 + peak2 + background
init = model.eval(pars, x=x)
out = model.fit(y, pars, x=x)
# print out.fit_report()
return init,out
def build_model_dd(bg_ord=1, peak_num=1, peak_type="lorentzian"):
model = PolynomialModel(bg_ord, prefix="bg_")
peak_str = peak_type.lower()[:2]
if peak_str == "lo":
peak_function = lorentzian_dd
elif peak_str == "ga":
peak_function = gaussian_dd
elif peak_str == "vo" or peak_type == "ps":
peak_function = voigt
else:
print(
"'peak_type' should be one of 'lorentzian', 'gaussian' or 'pseudo-voight'"
)
return
for i in range(peak_num):
model += Model(peak_function, prefix="p%s_" % i)
params = model.make_params()
return model, params
def build_model(self, peak_type='LO', max_width=None, bg_ord=2):
""" Builds a lmfit model of peaks in listed by index in `peak_pos`
Uses some basic algorithms to determine initial parameters for
amplitude and fwhm (limit on fwhm to avoid fitting background as peaks)
Parameters
----------
peak_type : string (default='lorentizian')
Peaks can be of the following types:
- 'LO' : symmetric lorentzian
- 'GA' : symmetric gaussain
max_width : int (default = total points/10)
max width (in data points) that peak fitted can be
bg_ord: int
order of the background polynomial
0: constant, 1: linear, ...
Returns
-------
pars : model parameters
model : model object
"""
x = self.x
y = self.y
pw = self.test_peak_width
peak_guess = self.x[self.peak_pos]
print("Building model ... ")
# start with polynomial background
# second order
model = PolynomialModel(bg_ord, prefix='bg_')
pars = model.make_params()
if peak_type == 'LO':
peak_function = lorentzian
self.afactor = pi
self.wfactor = 2.0
elif peak_type == 'GA':
peak_function = gaussian
self.afactor = sqrt(2 * pi)
self.wfactor = 2.354820
elif peak_type == 'VO':
peak_function = voigt
self.afactor = sqrt(2 * pi)
self.wfactor = 3.60131
# add lorentizian peak for all peaks
for i, peak in enumerate(peak_guess):
temp_model = Model(peak_function, prefix='p%s_' % i)
pars.update(temp_model.make_params())
model += temp_model
# set inital background as flat line at zeros
for i in range(bg_ord + 1):
pars['bg_c%i' % i].set(0)
# give values for other peaks
for i, peak in enumerate(self.peak_pos):
print('Peak %i: pos %s, height %s' % (i, x[peak], y[peak]))
# could set bounds #, min=x[peak]-5, max=x[peak]+5)
pars['p%s_center' % i].set(x[peak])
pars['p%s_sigma' % i].set(pw / 2, min=pw * 0.25, max=pw * 2)
# here as well #, min=0, max=2*max(y))
pars['p%s_amplitude' % i].set(self.amplitude(y[peak], (pw / 2)))
self.pars = pars
self.model = model
return self.pars, self.model
def fit_data(x, y, peak_pos, peak_type='LO', max_width=None, bg_ord=2):
""" Builds a lmfit model of peaks in listed by index in `peak_pos`
Uses some basic algorithms to determine initial parameters for
amplitude and fwhm (limit on fwhm to avoid fitting background as peaks)
Parameters
----------
peak_type : string (default='lorentizian')
Peaks can be of the following types:
- 'LO' : symmetric lorentzian
- 'GA' : symmetric gaussain
- 'VO' : symmetric pseudo voigt
max_width : int (default = total points/10)
max width (in data points) that peak fitted can be
bg_ord: int
order of the background polynomial
0: constant, 1: linear, ...
Returns
-------
pars : model parameters
model : model object
"""
# need to define peak width finding
pw = guess_peak_width(x, y)
peak_guess = x[peak_pos]
# start with polynomial background
model = PolynomialModel(bg_ord, prefix='bg_')
pars = model.make_params()
if peak_type == 'LO':
peak_function = lorentzian
elif peak_type == 'GA':
peak_function = gaussian
elif peak_type == 'VO':
peak_function = voigt
# add lorentizian peak for all peaks
for i, peak in enumerate(peak_guess):
temp_model = Model(peak_function, prefix='p%s_' % i)
pars.update(temp_model.make_params())
model += temp_model
# set inital background as flat line at zeros
for i in range(bg_ord + 1):
pars['bg_c%i' % i].set(0)
# give values for other peaks
for i, peak in enumerate(peak_pos):
# could set bounds #, min=x[peak]-5, max=x[peak]+5)
pars['p%s_x0' % i].set(x[peak])
pars['p%s_fwhm' % i].set(pw / 2, min=pw * 0.25, max=pw * 2)
# here as well #, min=0, max=2*max(y))
pars['p%s_amp' % i].set(y[peak])
out = model.fit(y, pars, x=x)
return out
def build_model(self, peak_type="LO", max_width=None, bg_ord=2):
""" Builds a lmfit model of peaks in listed by index in `peak_pos`
Uses some basic algorithms to determine initial parameters for
amplitude and fwhm (limit on fwhm to avoid fitting background as peaks)
Parameters
----------
peak_type : string (default='lorentizian')
Peaks can be of the following types:
- 'LO' : symmetric lorentzian
- 'GA' : symmetric gaussain
max_width : int (default = total points/10)
max width (in data points) that peak fitted can be
bg_ord: int
order of the background polynomial
0: constant, 1: linear, ...
Returns
-------
pars : model parameters
model : model object
"""
x = self.x
y = self.y
pw = self.test_peak_width
peak_guess = self.x[self.peak_pos]
print("Building model ... ")
# start with polynomial background
# second order
model = PolynomialModel(bg_ord, prefix="bg_")
pars = model.make_params()
if peak_type == "LO":
peak_function = lorentzian
self.afactor = pi
self.wfactor = 2.0
elif peak_type == "GA":
peak_function = gaussian
self.afactor = sqrt(2 * pi)
self.wfactor = 2.354820
elif peak_type == "VO":
peak_function = voigt
self.afactor = sqrt(2 * pi)
self.wfactor = 3.60131
# add lorentizian peak for all peaks
for i, peak in enumerate(peak_guess):
temp_model = Model(peak_function, prefix="p%s_" % i)
pars.update(temp_model.make_params())
model += temp_model
# set inital background as flat line at zeros
for i in range(bg_ord + 1):
pars["bg_c%i" % i].set(0)
# give values for other peaks
for i, peak in enumerate(self.peak_pos):
print("Peak %i: pos %s, height %s" % (i, x[peak], y[peak]))
# could set bounds #, min=x[peak]-5, max=x[peak]+5)
pars["p%s_center" % i].set(x[peak])
pars["p%s_sigma" % i].set(pw / 2, min=pw * 0.25, max=pw * 2)
# here as well #, min=0, max=2*max(y))
pars["p%s_amplitude" % i].set(self.amplitude(y[peak], (pw / 2)))
self.pars = pars
self.model = model
return self.pars, self.model
def fit(spectra, obj, sigma=2.0, ord=4, iter=4):
poly = PolynomialModel(3)
pars = poly.make_params()
for p in range(4):
label = 'c'+str(p)
pars[label].set(value=1., vary=True)
wkcopy = np.copy(spectra[1])
truesp = [i for i in wkcopy if i > 5]
truex = [spectra[0][i] for i in range(len(spectra[1])) if spectra[1][i] > 5]
outcont = poly.fit(truesp, pars, x=truex)
firstcont = outcont.eval(x=spectra[0])
xn = np.copy(spectra[0])
yn = np.copy(spectra[1])/firstcont
pl1=plt.subplot((iter+1)*100+11)
pl1.plot(xn, spectra[1], 'k-', linewidth=0.3)
pl1.plot(xn, firstcont, 'r-', linewidth=0.6)
pl1.set_ylim([0, np.mean(firstcont)*1.5])
for i in range(iter):
i_=np.copy(i)
niter=str(i_+1)
sigma = sigma-i*0.21*sigma
md = np.median(yn)
n = len([i for i in yn if i > 0.1])
offset = (len(xn)-n)/2
absor = md - min(yn[offset:n-offset])
freq, bin = np.histogram(yn, bins=50, range=(md-absor, md+absor))
rebin = [(bin[b+1]+bin[b])/2 for b in range(len(bin)-1)]
gauss = SkewedGaussianModel()
pars = gauss.make_params()
pars['center'].set(value=md, vary=True)
pars['amplitude'].set(vary=True)
pars['sigma'].set(vary=True)
pars['gamma'].set(vary=True)
out = gauss.fit(freq, pars, x=rebin)
var = sigma*out.best_values['sigma']
xrbn = np.linspace(rebin[0], rebin[-1], num=100)
yrbn = list(out.eval(x=xrbn))
mode = xrbn[yrbn.index(max(yrbn))]
ync = np.copy(spectra[1])
xnc = np.copy(spectra[0])
mask = []
for j in range(len(yn)):
if (yn[j] > mode+var/2) or (yn[j] < mode-var/2):
mask.append(False)
else:
mask.append(True)
mask = np.array(mask)
ync = ync[mask]
xnc = xnc[mask]
poly2 = PolynomialModel(ord)
pars2 = poly2.make_params()
for p in range(ord+1):
label = 'c'+str(p)
pars2[label].set(value=1., vary=True)
outcont2 = poly2.fit(ync, pars2, x=xnc)
contf = outcont2.eval(x=xn)
yn = spectra[1]/contf
err = spectra[2]/contf
pln=plt.subplot(int((iter+1)*100+10+(i_+2)))
pln.plot(xn, yn*(np.mean(contf)*0.8), 'k-', linewidth=0.3)
pln.plot(xnc, ync, 'r-', linewidth=0.3)
pln.plot(xn, contf, 'b-', linewidth=0.6)
pln.set_ylim([0, np.mean(contf)*1.2])
plt.savefig(obj[0]+'_fit.png', dpi=300)
plt.clf()
return np.array([xn, yn, err])
def fit(spectra, sigma=4.0, ptreg=8., ord=4, iter=2):
rej = []
for i, w in enumerate(spectra[1]):
if w <= 0:
rej.append(i)
spectra[0] = np.delete(spectra[0], rej)
spectra[1] = np.delete(spectra[1], rej)
# prepare first kick
poly = PolynomialModel(3)
pars = poly.make_params()
for p in range(4):
label = 'c' + str(p)
pars[label].set(value=1., vary=True)
wkcopy = np.copy(spectra[1])
truesp = [i for i in wkcopy if i >= 0]
truex = [
spectra[0][i] for i in range(len(spectra[1])) if spectra[1][i] >= 0
]
outcont = poly.fit(truesp, pars, x=truex)
firstcont = outcont.eval(x=spectra[0])
xn = np.copy(spectra[0])
yn = np.copy(spectra[1]) / firstcont
# start cont. cleaning iterations
for i in range(iter):
i_ = np.copy(i)
niter = str(i_ + 1)
sigma = sigma - i * 0.21 * sigma
md = np.median(yn)
n = len([i for i in yn if i > 0.1])
offset = (len(xn) - n) / 2
absor = md - min(yn[offset:n - offset])
freq, bin = np.histogram(yn, bins=50, range=(md - absor, md + absor))
rebin = [(bin[b + 1] + bin[b]) / 2 for b in range(len(bin) - 1)]
gauss = SkewedGaussianModel()
pars = gauss.make_params()
pars['center'].set(vary=True)
pars['amplitude'].set(vary=True)
pars['sigma'].set(vary=True)
pars['gamma'].set(vary=True)
out = gauss.fit(freq, pars, x=rebin)
var = sigma * out.best_values['sigma']
xrbn = np.linspace(rebin[0], rebin[-1], num=100)
yrbn = list(out.eval(x=xrbn))
mode = xrbn[yrbn.index(max(yrbn))]
# clean cont.
ync = np.copy(spectra[1])
xnc = np.copy(spectra[0])
mask = []
for j in range(len(yn)):
if (yn[j] > mode + var / 2) or (yn[j] < mode - var / 2):
mask.append(False)
else:
mask.append(True)
mask = np.array(mask)
ync = ync[mask]
xnc = xnc[mask]
# re-fitting
poly2 = PolynomialModel(ord)
pars2 = poly2.make_params()
for p in range(ord + 1):
label = 'c' + str(p)
pars2[label].set(value=1., vary=True)
try:
outcont2 = poly2.fit(ync, pars2, x=xnc)
except:
plt.plot(xn, yn, 'k-')
plt.plot([xn[0], xn[-1]], [mode, mode], 'b-')
plt.plot([xn[0], xn[-1]], [mode + var / 2, mode + var / 2], 'r-')
plt.plot([xn[0], xn[-1]], [mode - var / 2, mode - var / 2], 'r-')
plt.show()
contf = outcont2.eval(x=xn)
yn = spectra[1] / contf
clspec = [xnc, ync]
# start slicing
firstv = clspec[0][0]
wavrange = clspec[0][-1] - firstv
sliceno = wavrange / ptreg
slisize = wavrange / sliceno
points = [[], []]
# continuum point definition
for s in range(int(sliceno)):
i = bissec(clspec[0], firstv + s * slisize)
f = bissec(clspec[0], firstv + (s + 1) * slisize)
slc = [clspec[0][i:f], clspec[1][i:f]]
if len(slc[1]) > 2.:
md = np.median(slc[1])
absor = min(slc[1])
high = max(slc[1])
freq, bin = np.histogram(slc[1], bins=20, range=(absor, high))
rebin = [(bin[b + 1] + bin[b]) / 2 for b in range(len(bin) - 1)]
fmode = rebin[list(freq).index(max(freq))]
fsigma = rebin[-1] - rebin[0]
gauss = GaussianModel()
pars = gauss.make_params()
pars['center'].set(value=fmode, vary=True)
pars['amplitude'].set(value=max(freq), vary=True)
pars['sigma'].set(value=fsigma, vary=True)
out = gauss.fit(freq, pars, x=rebin)
xrbn = np.linspace(rebin[0], rebin[-1], num=100)
yrbn = list(out.eval(x=xrbn))
mode = xrbn[yrbn.index(max(yrbn))]
xp = slc[0][len(slc[0]) / 2]
points[0].append(xp)
points[1].append(mode)
spline = splrep(points[0], points[1], k=3)
contx = splev(clspec[0], spline)
continuum = splev(spectra[0], spline)
return [spectra[0], spectra[1] / continuum]
