def fit_gaussian(data, constrained=False):
illusion_strength = compute_illusion_strength(data)
noise_frequencies = np.unique(data.noise_freq)
n_reps = illusion_strength.shape[1]
# fit a horizontal line if the data do not have any real dip
# approximate baseline as 3rd largest value
baseline = np.sort(illusion_strength.mean(1))[-2]
minimum = illusion_strength.mean(1).min()
if baseline - minimum < illusion_strength.mean(1).max() - baseline:
model = lmfit.Model(lambda x, baseline: baseline)
params = lmfit.Parameters()
params.add('baseline', value=baseline)
else:
params = lmfit.Parameters()
#params.add('center', value=3)
params.add('center',
value=noise_frequencies[np.argmin(
illusion_strength.mean(1))],
max=9,
min=.5)
params.add('width', value=2)
if constrained:
params.add('baseline', value=baseline)
params.add('minimum',
value=minimum,
min=minimum - 1,
max=baseline + 1)
else:
params.add('baseline', value=5)
params.add('minimum', value=-2)
model = lmfit.Model(inverted_gaussian)
result = model.fit(illusion_strength.ravel(),
x=noise_frequencies.repeat(n_reps),
params=params)
return result
def fit_gaussian(data, constrained=False):
illusion_strength = compute_illusion_strength(data)
noise_frequencies = np.unique(data.noise_freq)
n_reps = illusion_strength.shape[1]
# fit a horizontal line if the data do not have any real dip
if data.vp[0] in ['dbm', 'odog'] or (data.vp[0] == 'biwam'
and data.grating_freq[0] == .2):
model = lmfit.Model(lambda x, baseline: baseline)
params = lmfit.Parameters()
params.add('baseline', value=5)
else:
params = lmfit.Parameters()
#params.add('center', value=3)
params.add('center',
value=noise_frequencies[np.argmin(
illusion_strength.mean(1))],
max=9,
min=.01)
params.add('width', value=3)
if constrained:
minimum = illusion_strength.mean(1).min()
params.add('baseline', value=5, max=8.8)
params.add('minimum', value=-minimum, min=-6, max=-minimum + 1)
else:
params.add('baseline', value=5)
params.add('minimum', value=5)
model = lmfit.Model(inverted_gaussian)
result = model.fit(illusion_strength.ravel(),
x=noise_frequencies.repeat(n_reps),
params=params)
return result
def __init__(self, amplitude, sigma, skew, baseline, floor, ceiling, null, peak_limit_frc=.2):
# initialise null model
null_model = lmfit.models.ConstantModel()
null_params = lmfit.Parameters()
null_params['c'] = null
# initialise peak model
peak_model = lmfit.Model(skewed_gaussian)
peak_params = lmfit.Parameters()
peak_params['center'] = lmfit.Parameter(value=0, vary=False)
peak_params['amplitude'] = amplitude
peak_params['sigma'] = sigma
peak_params['skew'] = skew
peak_params['baseline'] = baseline
peak_params['ceiling'] = ceiling
peak_params['floor'] = floor
# initialise flank model
half_peak_model = lmfit.Model(gaussian)
super(SkewedGaussianPeakFitter, self).__init__(
null_model=null_model, null_params=null_params, peak_model=peak_model,
peak_params=peak_params, half_peak_model=half_peak_model, peak_limit_frc=peak_limit_frc
)
def prepare_fitting(self):
self.fit_dicts = OrderedDict()
dac_vals = self.proc_data_dict['dac_values']
freq_vals = self.proc_data_dict['fit_frequencies']
if max(freq_vals) < 1e9:
freq_vals *= 1e9
guess = self.options_dict.get('fit_guess', {})
f_q = guess.get(
'f_max_qubit', self.options_dict.get('f_max_qubit', None))
guess['f_max_qubit'] = f_q
ext = f_q is not None
if self.is_spectroscopy:
fitmod = lmfit.Model(Qubit_dac_to_freq)
fitmod.guess = Qubit_dac_arch_guess.__get__(
fitmod, fitmod.__class__)
else:
if f_q is None and self.verbose:
print('Specify f_max_qubit in the options_dict to obtain a better fit!')
# Todo: provide alternative fit?
fitmod = lmfit.Model(Resonator_dac_to_freq)
fitmod.guess = Resonator_dac_arch_guess.__get__(
fitmod, fitmod.__class__)
# fit_result = fitmod.fit(freq_vals, dac_voltage=dac_vals)
self.fit_dicts['dac_arc'] = {
'model': fitmod,
'fit_xvals': {'dac_voltage': dac_vals},
'fit_yvals': {'data': freq_vals},
'guessfn_pars': {'values': guess}
}
def curv_fit(x=None, y=None, model=None):
x = np.array(x)
y = np.array(y)
params = lmfit.Parameters()
if model == 'gaussian':
mod = lmfit.models.GaussianModel()
params = mod.guess(y, x=x)
out = mod.fit(y, params, x=x)
r_sq = 1 - out.residual.var() / np.var(y)
elif model == '4PL':
mod = lmfit.Model(logistic_4p)
params.add('la', value=1.0)
params.add('gr', value=120.0, vary=False)
params.add('ce', value=150.0)
params.add('ua', value=3.0)
out = mod.fit(y, params, x=x)
r_sq = 1 - out.residual.var() / np.var(y)
elif model == '5PL':
mod = lmfit.Model(logistic_5p)
params.add('la', value=1.0)
params.add('gr', value=1.0)
params.add('ce', value=1.0)
params.add('ua', value=1.0)
params.add('sy', value=1.0)
out = mod.fit(y, params, x=x)
r_sq = 1 - out.residual.var() / np.var(y)
out.R_sq = r_sq
return out
def lmfit_voigt_fit(data, startparams, costrains):
# startparams:XCEN,YCEN,
model = lmfit.Model(func=constant_func,
independent_vars=['x', 'y'],
prefix='const_')
for i in range(len(startparams)):
mod = lmfit.Model(func=pseudo_voigt_2d,
independent_vars=['x', 'y'],
prefix='model' + str(i) + '_')
model += mod
params = model.make_params()
params['const_c'].set(value=np.min(data[2]), vary=False)
for i in range(len(startparams)):
for j in values:
ii = 'model' + str(i) + '_'
params[ii + j].set(value=startparams.at[ii, j],
max=costrains.at[j, 'max'],
min=costrains.at[j, 'min'],
vary=costrains.at[j, 'vary'])
output = model.fit(data=data[2].ravel(),
params=params,
x=data[0].ravel(),
y=data[1].ravel())
return output
def Obtain_Initial_Phase(tpf, corrected_lc, frequency_list):
flux = corrected_lc.flux.value
times = corrected_lc.time.value - np.mean(corrected_lc.time.value)
pg = corrected_lc.to_periodogram(frequency=np.append(
[0.0001], frequency_list),
ls_method='slow')
initial_flux = np.asarray(pg.power[1:])
initial_phase = np.zeros(len(frequency_list))
def lc_model(time, amp, freq, phase):
return amp * np.sin(2 * np.pi * freq * time + phase)
def background_model(time, height):
return np.ones(len(time)) * height
for j in np.arange(len(frequency_list)):
for i in np.arange(len(frequency_list)):
if (i == 0):
model = lm.Model(lc_model,
independent_vars=['time'],
prefix='f{0:d}'.format(i))
model += lm.Model(background_model,
independent_vars=['time'])
else:
model += lm.Model(lc_model,
independent_vars=['time'],
prefix='f{0:d}'.format(i))
model.set_param_hint('f{0:d}phase'.format(i),
min=-np.pi,
max=np.pi,
value=initial_phase[i],
vary=False)
model.set_param_hint('f{0:d}amp'.format(i),
value=initial_flux[i],
vary=False)
model.set_param_hint('height',
value=np.mean(flux),
vary=False)
model.set_param_hint('f{0:d}freq'.format(i),
value=frequency_list[i],
vary=False)
params = model.make_params()
params['f{0:d}phase'.format(j)].set(vary=True)
params['f{0:d}phase'.format(j)].set(value=initial_phase[j])
params['f{0:d}phase'.format(j)].set(brute_step=np.pi / 10)
result = model.fit(corrected_lc.flux.value,
params,
time=times,
method='brute')
initial_phase[j] = result.best_values['f{0:d}phase'.format(j)]
return initial_phase
def __init__(self, amp, tau):
self.amp_s = amp
self.tau_s = tau
self.amp_k = self.amp_s/(self.amp_s-1.)
self.tau_k = (1.-self.amp_s)*self.tau_s
self.step_func = lambda t: 1.-self.amp_s*np.exp(-t/self.tau_s)
self.kernel_step_func = lambda t: 1.+self.amp_k*np.exp(-t/self.tau_k)
self.step_model = lmfit.Model(self.step_func)
self.kernel_step_model = lmfit.Model(self.kernel_step_func)
def fitBandy2(et, M, dM=None, mode="logM", start=None, lb=None, ub=None):
def residual(pars, x, data=None, eps=None):
# unpack parameters:
# extract .value attribute for each parameter
parvals = pars.valuesdict()
period = parvals["period"]
shift = parvals["shift"]
decay = parvals["decay"]
if abs(shift) > pi / 2:
shift = shift - sign(shift) * pi
if abs(period) < 1.0e-10:
period = sign(period) * 1.0e-10
model = parvals["amp"] * sin(shift + x / period) * exp(
-x * x * decay * decay)
if data is None:
return model
if eps is None:
return model - data
return (model - data) / eps
pars = lmfit.Parameters()
pars.add("t", value=start[0], min=lb[0], max=ub[0], vary=1)
pars.add("g", value=start[1], min=lb[1], max=ub[1], vary=1)
if dM is not None:
wgt = 1.0 / dM**2
else:
wgt = np.ones_like(M)
if mode == "M":
mod = lmfit.Model(bandy_M)
elif mode == "logM":
mod = lmfit.Model(bandy_logM)
elif mode == "logMa" or mode == "logMa_off":
mod = lmfit.Model(bandy_logMa)
if "off" in mode:
pars.add("a", value=0, min=0, vary=1)
else:
pars.add("a", value=0, vary=0)
pars.add("b", value=start[2], min=lb[2], max=ub[2], vary=1)
elif mode == "logV2a":
mod = lmfit.Model(bandy_V2a)
wgt = np.log10(wgt)
pars.add("a", value=0, vary=0)
pars.add("b", value=start[2], min=lb[2], max=ub[2], vary=1)
if "log" in mode:
M = np.log10(M)
if dM is not None:
wgt = 1.0 / np.log10(dM)**2
return mod.fit(M, pars, x=et, weights=wgt)
def fitBandy2(et, M, dM=None, mode='logM', start=None, lb=None, ub=None):
def residual(pars, x, data=None, eps=None):
# unpack parameters:
# extract .value attribute for each parameter
parvals = pars.valuesdict()
period = parvals['period']
shift = parvals['shift']
decay = parvals['decay']
if abs(shift) > pi / 2:
shift = shift - sign(shift) * pi
if abs(period) < 1.e-10:
period = sign(period) * 1.e-10
model = parvals['amp'] * sin(shift + x / period) * exp(
-x * x * decay * decay)
if data is None:
return model
if eps is None:
return (model - data)
return (model - data) / eps
pars = lmfit.Parameters()
pars.add('t', value=start[0], min=lb[0], max=ub[0], vary=1)
pars.add('g', value=start[1], min=lb[1], max=ub[1], vary=1)
if dM is not None:
wgt = 1. / dM**2
else:
wgt = np.ones_like(M)
if mode == 'M':
mod = lmfit.Model(bandy_M)
elif mode == 'logM':
mod = lmfit.Model(bandy_logM)
elif mode == 'logMa' or mode == 'logMa_off':
mod = lmfit.Model(bandy_logMa)
if 'off' in mode:
pars.add('a', value=0, min=0, vary=1)
else:
pars.add('a', value=0, vary=0)
pars.add('b', value=start[2], min=lb[2], max=ub[2], vary=1)
elif mode == 'logV2a':
mod = lmfit.Model(bandy_V2a)
wgt = np.log10(wgt)
pars.add('a', value=0, vary=0)
pars.add('b', value=start[2], min=lb[2], max=ub[2], vary=1)
if 'log' in mode:
M = np.log10(M)
if dM is not None:
wgt = (1. / np.log10(dM)**2)
return mod.fit(M, pars, x=et, weights=wgt)
def get_optimal_amp(qbc, qbt, soft_sweep_points, timestamp=None,
classified_ro=False, tangent_fit=False,
parfit=False,
analysis_object=None, **kw):
if analysis_object is None:
if classified_ro:
channel_map = {qb.name: [vn + ' ' +
qb.instr_acq() for vn in
qb.int_avg_classif_det.value_names]
for qb in [qbc, qbt]}
else:
channel_map = {qb.name: [vn + ' ' +
qb.acq_instr() for vn in
qb.int_avg_det.value_names]
for qb in [qbc, qbt]}
tdma = tda.CPhaseLeakageAnalysis(
t_start=timestamp,
qb_names=[qbc.name, qbt.name],
options_dict={'TwoD': True, 'plot_all_traces': False,
'plot_all_probs': False,
'delegate_plotting': False,
'channel_map': channel_map})
else:
tdma = analysis_object
cphases = tdma.proc_data_dict[
'analysis_params_dict'][f'cphase_{qbt.name}']['val']
sweep_pts = list(soft_sweep_points.values())[0]['values']
if tangent_fit:
fit_res = lmfit.Model(lambda x, m, b: m*np.tan(x/2-np.pi/2) + b).fit(
x=cphases, data=sweep_pts,
m=(max(sweep_pts)-min(sweep_pts))/((max(cphases)-min(cphases))),
b=np.min(sweep_pts))
elif parfit:
fit_res = lmfit.Model(lambda x, m, b, c: m*x + c*x**2 + b).fit(
x=cphases, data=sweep_pts,
m=(max(sweep_pts)-min(sweep_pts))/((max(cphases)-min(cphases))),
c=0.001,
b=np.min(sweep_pts))
else:
fit_res = lmfit.Model(lambda x, m, b: m*x + b).fit(
x=cphases, data=sweep_pts,
m=(max(sweep_pts)-min(sweep_pts))/((max(cphases)-min(cphases))),
b=np.min(sweep_pts))
plot_and_save_cz_amp_sweep(cphases=cphases, timestamp=timestamp,
soft_sweep_params_dict=soft_sweep_points,
fit_res=fit_res, save_fig=True, plot_guess=False,
qbc_name=qbc.name, qbt_name=qbt.name, **kw)
return fit_res
def test_live_fit_plot(fresh_RE):
RE = fresh_RE
try:
import lmfit
except ImportError:
raise pytest.skip('requires lmfit')
def gaussian(x, A, sigma, x0):
return A * np.exp(-(x - x0)**2 / (2 * sigma**2))
model = lmfit.Model(gaussian)
init_guess = {
'A': 2,
'sigma': lmfit.Parameter('sigma', 3, min=0),
'x0': -0.2
}
livefit = LiveFit(model,
'det', {'x': 'motor'},
init_guess,
update_every=50)
lfplot = LiveFitPlot(livefit, color='r')
lplot = LivePlot('det', 'motor', ax=plt.gca(), marker='o', ls='none')
RE(scan([det], motor, -1, 1, 50), [lplot, lfplot])
expected = {'A': 1, 'sigma': 1, 'x0': 0}
for k, v in expected.items():
assert np.allclose(livefit.result.values[k], v, atol=1e-6)
def test_live_fit_multidim(fresh_RE):
RE = fresh_RE
try:
import lmfit
except ImportError:
raise pytest.skip('requires lmfit')
motor1._fake_sleep = 0
motor2._fake_sleep = 0
det4.exposure_time = 0
def gaussian(x, y, A, sigma, x0, y0):
return A * np.exp(-((x - x0)**2 + (y - y0)**2) / (2 * sigma**2))
model = lmfit.Model(gaussian, ['x', 'y'])
init_guess = {
'A': 2,
'sigma': lmfit.Parameter('sigma', 3, min=0),
'x0': -0.2,
'y0': 0.3
}
cb = LiveFit(model,
'det4', {
'x': 'motor1',
'y': 'motor2'
},
init_guess,
update_every=50)
RE(outer_product_scan([det4], motor1, -1, 1, 10, motor2, -1, 1, 10, False),
cb)
expected = {'A': 1, 'sigma': 1, 'x0': 0, 'y0': 0}
for k, v in expected.items():
assert np.allclose(cb.result.values[k], v, atol=1e-6)
def test_live_fit_plot(RE, hw):
try:
import lmfit
except ImportError:
raise pytest.skip("requires lmfit")
def gaussian(x, A, sigma, x0):
return A * np.exp(-(x - x0)**2 / (2 * sigma**2))
model = lmfit.Model(gaussian)
init_guess = {
"A": 2,
"sigma": lmfit.Parameter("sigma", 3, min=0),
"x0": -0.2,
}
livefit = LiveFit(model,
"det", {"x": "motor"},
init_guess,
update_every=50)
lfplot = LiveFitPlot(livefit, color="r")
lplot = LivePlot("det", "motor", ax=plt.gca(), marker="o", ls="none")
RE(scan([hw.det], hw.motor, -1, 1, 50), [lplot, lfplot])
expected = {"A": 1, "sigma": 1, "x0": 0}
for k, v in expected.items():
assert np.allclose(livefit.result.values[k], v, atol=1e-6)
def train():
body = request.json
df=pd.DataFrame(body)
N=body["init"].N
#data=np.array([body["confirmed"],body["active"],body["recovered"],body["deaths"]])
data=np.array([body["confirmed"],body["active"]])
x =df.index
# plt.plot(x,data[1])
# plt.show()
print(len(data.T),len(x))
mod = lmfit.Model(mdl)
mod.set_param_hint("beta", value=body["init"].beta, vary=True,min=0, max=6)
mod.set_param_hint("gamma", value=body["init"].gamma, vary=True,min=0, max=4,)
mod.set_param_hint("sigma", value=body["init"].sigma, vary=True,min=0, max=4)
mod.set_param_hint("N", value=N, vary=False)
params = mod.make_params()
result = mod.fit(data, params, method="leastsq", x=x)
#result.plot_fit(datafmt="-")
#plt.show()
print(result.fit_report())
# a=requests.get(apiUrl+"dayone/country/"+body["country"])
# print(a.content)
return dict(result.params.valuesdict())
def test_coercion_of_input_data(peakdata, input_dtype, expected_dtype):
"""Test for coercion of 'data' and 'independent_vars'.
- 'data' should become 'float64' or 'complex128'
- dtype for 'indepdendent_vars' is only changed when the input is a list,
tuple, numpy.ndarray, or pandas.Series
"""
x, y = peakdata
model = lmfit.Model(gaussian)
pars = model.make_params()
if (not lmfit.minimizer.HAS_PANDAS and input_dtype in ['pandas-real',
'pandas-complex']):
return
elif input_dtype == 'pandas-real':
result = model.fit(lmfit.model.Series(y, dtype=np.float32), pars,
x=lmfit.model.Series(x, dtype=np.float32))
elif input_dtype == 'pandas-complex':
result = model.fit(lmfit.model.Series(y, dtype=np.complex64), pars,
x=lmfit.model.Series(x, dtype=np.complex64))
elif input_dtype == 'list':
result = model.fit(y.tolist(), pars, x=x.tolist())
elif input_dtype == 'tuple':
result = model.fit(tuple(y), pars, x=tuple(x))
else:
result = model.fit(np.asarray(y, dtype=input_dtype), pars,
x=np.asarray(x, dtype=input_dtype))
assert result.__dict__['userkws']['x'].dtype == expected_dtype
assert result.__dict__['userargs'][0].dtype == expected_dtype
def fit(self):
mod = lmfit.Model(self.fitter)
self.kabko.apply_params(mod)
outbreak_shift = self.kabko.outbreak_shift(
self.kabko.params["incubation_period"].init
)
days = self.kabko.data_days(outbreak_shift)
mod.set_param_hint("days", value=days, vary=False)
params = mod.make_params()
x_data = np.linspace(0, days - 1, days, dtype=int)
y_data = util.shift_array(
self.kabko.dead,
outbreak_shift
)
result = mod.fit(y_data, params, method="least_squares", x=x_data)
#result.plot_fit(datafmt="-");
return result
def tip_line_fit(point1, point2, image, length_spacing, line_thickness, width_spacing):
"""Fit the tip of a line to a complementary error function, 1 - erf(x).
Args:
point1: 1D array or float.
point2: 1D array or float.
image: 2D array.
length_spacing: float, `get_thick_line()`.
line_thickness: float, `get_thick_line()`.
width_spacing: float, `get_thick_line()`.
"""
profile_parameters = {}
profile_parameters["length_spacing"] = length_spacing # pixel
profile_parameters["line_thickness"] = line_thickness # pixel
profile_parameters["width_spacing"] = width_spacing # pixel
profile_parameters["normalized_intensities"] = True
x_profile, y_profile = line_profile(image, point1, point2, **profile_parameters)
def errorfunction(x, mu, sigma, mt, bg):
# pylint: disable=no-member
return bg + (0.5 * mt * special.erfc((x - mu) / (np.sqrt(2) * sigma)))
model = lmfit.Model(errorfunction)
fit_params = {}
fit_params["mu"] = lmfit.Parameter("mu", value=x_profile[-1] / 2, min=0, max=x_profile[-1])
fit_params["sigma"] = lmfit.Parameter("sigma", value=1, vary=True, min=0, max=x_profile[-1])
fit_params["mt"] = lmfit.Parameter("mt", value=1, vary=True, min=0)
fit_params["bg"] = lmfit.Parameter("bg", value=1, vary=True, min=0)
fit_result = model.fit(y_profile.copy(), x=x_profile.copy(), **fit_params)
return x_profile, y_profile, fit_result, errorfunction
def setup():
fig, ax = plt.subplots()
self.ax = ax
fig.canvas.set_window_title('Peakup')
self.ax.clear()
# Setup LiveCallbacks
self.live_plot = LivePlot(scaler,
dcm_c2_pitch_name,
linestyle='',
marker='*',
color='C0',
label='raw',
ax=self.ax,
use_teleporter=False)
# Setup LiveFit
# f_gauss(x, A, sigma, x0, y0, m)
model = lmfit.Model(f_gauss, ['x'])
init_guess = {
'A': lmfit.Parameter('A', 100000, min=0),
'sigma': lmfit.Parameter('sigma', 0.001, min=0),
'x0': lmfit.Parameter('x0', pitch_guess),
'y0': lmfit.Parameter('y0', 0, min=0),
'm': lmfit.Parameter('m', 0, vary=False)
}
self.lf = LiveFit(model, scaler, {'x': dcm_c2_pitch_name},
init_guess)
self.lpf = LiveFitPlot(self.lf,
ax=self.ax,
color='r',
use_teleporter=False,
label='Gaussian fit')
def __init__(self, n_peaks, verbose=False, model='Lorentz'):
self.mkey = model
_func = {
'Lorentz': makefunc_Lorentz_cmplx,
'Voigt': makefunc_Voigt
}[model]
self.n = n_peaks
self.model = lmfit.Model(_func(n_peaks))
self.params = self.model.make_params(verbose=False)
# This loop initializes all parameters by 1 and sets standard boundaries depending on the respective parameterset
for k in self.params.keys():
self.set_initial_values(k, 1)
if k[0] == 'a':
self.set_boundaries(k, 0, np.inf)
self.set_initial_values(k, 1e6)
elif k[0] == 'l':
self.set_boundaries(k, 0, np.inf)
self.set_initial_values(k, 1e-2)
elif k[0] == 'p':
self.set_boundaries(k, -180, 180)
elif k[0] == 's':
self.set_boundaries(k, 0, np.inf)
self.set_initial_values(k, 1e-2)
else:
pass
if verbose:
self.params.pretty_print()
def simplefit(t, G, err, kappa = 5, triplet = 1):
#0 = triplet off, 1 = triplet on
# 1-comp diffusion, triplet considered only if selected
A0_guess = np.mean(G[:5])
arg = abs(G - A0_guess/2)
ind = np.argmin(arg)
tau_guess = t[ind]
print('Guessing tau = %f ms' %(tau_guess))
model = lmfit.Model(ff.triplet_diff3d)
params = model.make_params(A0=A0_guess, tau_diff=tau_guess)
if triplet == 1:
params['T'].set(min=0, max=1, value=0.1)
params['tau_trip'].set(min=1E-9, max=0.05, value=10e-4)
else:
params['T'].set(value=0, vary=False)
params['tau_trip'].set(value=1e-4, vary=False)
params['A0'].set(min=0.00001, value=A0_guess)
params['tau_diff'].set(min=1e-4, value=tau_guess)
params['Ginf'].set(value=0, vary = True)
params['kappa'].set(value=kappa, vary=False) # 3D model only
weights = 1/err
t_max = 1e7
fitres = model.fit(G[t < t_max], timelag=t[t<t_max], params=params, method='least_squares',
weights=weights[t<t_max])
print('\nList of fitted parameters for %s: \n' % model.name)
fitres.params.pretty_print(colwidth=10, columns=['value', 'stderr', 'min', 'max'])
print(fitres.fit_report())
return fitres
def simplefit2(t, G, err, kappa = 5, tau_diff2_fix_value = 0.050):
# tau_diff2_fix_value = 0.036
# kappa = 5.9
# two component diffusion
model = lmfit.Model(ff.twocomp_diff3d)
A0_guess = np.mean(G[:5])
arg = abs(G - A0_guess/2)
ind = np.argmin(arg)
tau_guess = t[ind]
print('Guessing tau = %f ms' %(tau_guess))
params = model.make_params(A0=A0_guess)
params['A0'].set(min=0.00001, value = A0_guess)
# params['tau_diff1'].set(min=2*tau_diff2_fix_value, value = tau_guess) #slow component
params['tau_diff1'].set(min=1e-6, value = tau_guess) #slow component
params['tau_diff2'].set(min=1e-6, value = tau_diff2_fix_value, vary = False) #fast component, usually fixed
params['p1'].set(min = 0, max = 1, value = 0.5) #fraction slow
params['Ginf'].set(value=0, vary = True)
params['kappa'].set(value=kappa, vary=False) # 3D model only
# weights = np.ones_like(avg_G)
weights = 1/err
t_max = 1e7
fitres = model.fit(G[t < t_max], timelag=t[t<t_max], params=params, method='least_squares',
weights=weights[t<t_max])
print('\nList of fitted parameters for %s: \n' % model.name)
fitres.params.pretty_print(colwidth=10, columns=['value', 'stderr', 'min', 'max'])
print(fitres.fit_report())
return fitres
def constrained_fitter(minicube, xax, input_parameters, par_minima=None,
par_maxima=None, **model_kwargs):
"""
input_parameters should be a dict
"""
model = lmfit.Model(multicomp_minicube_model_generator(**model_kwargs),
independent_vars=['xax'])
params = model.make_params()
for par in params:
params[par].value = input_parameters[par]
if 'amp' in par and par[3] != 'd':
params[par].min = 0
elif 'sigma' in par and par[5] != 'd':
params[par].min = 0
if par_minima is not None:
for mpar in par_minima:
if mpar not in params:
raise ValueError("Parameter {0} is not in params".format(mpar))
params[mpar].min = par_minima[mpar]
if par_maxima is not None:
for mpar in par_maxima:
if mpar not in params:
raise ValueError("Parameter {0} is not in params".format(mpar))
params[mpar].max = par_maxima[mpar]
result = model.fit(minicube, xax=xax,
params=params)
return result
def fit_to_exp(self, rem = 0, plot_rem = False):
model = lmf.Model(self.decay_exp)
params = model.make_params(amp = 1.0, delta = 0.5)
params.add('c', value = 0, vary = False)
# res_exp = model.fit(self.Fs, params, x = self.Ls)
mod = lmf.models.LinearModel()
pars = mod.guess(np.log(self.Fs[:self.Len-rem]), x=self.Ls[:self.Len-rem])
scaling_fit = mod.fit(np.log(self.Fs)[:self.Len-rem] ,pars, x=self.Ls[:self.Len-rem])
fig, ax = plt.subplots(figsize = (6, 3.6))
ax.plot(self.Ls, self.Fs, "o")
ax.plot(self.Ls, np.exp(scaling_fit.values["intercept"])*np.exp(scaling_fit.values["slope"]*self.Ls),"C1-")
ax.annotate("$\propto exp({:.2f}{})$".format(scaling_fit.values["slope"], self.L_label), (self.Ls[0]*0.1+self.Ls[1]*0.9, self.Fs[0]*0.1 + self.Fs[1]*0.9))
ax.set_xlabel("{}".format(self.L_label))
ax.set_ylabel("{}".format(self.F_label))
ax.tick_params(axis = "y")
ax.yaxis.set_major_formatter(mpl.ticker.FormatStrFormatter('%.1e'))
ax.locator_params(nbins=5)
axin = ax.inset_axes([0.35,0.35,0.6,0.6])
scaling_fit.plot_fit(ax = axin, xlabel = "{}".format(self.L_label), ylabel = "log({})".format(self.F_label))
if plot_rem:
axin.plot(self.Ls[-3:],np.log(np.abs(self.Fs[-rem:])),"C0o")
axin.set_title("")
axin.set_xticklabels([])
axin.set_yticklabels([])
axin.tick_params(axis='both', which='both', length=0)
axin.get_legend().remove()
plt.show()
def read_mu_x(self):
self.debug_stream("Calculating the center of the peak in the x axis")
def gaussian_paula(x, mu, A, sigma, c):
return A * np.exp(-(x - mu)**2 / (2 * sigma**2)) + c
mod = lm.Model(gaussian_paula)
self.parsx = lm.Parameters()
real_data = np.array(self.image_proxy.read().value)
self.x_axis = np.mean(real_data, axis=0)
#self.x_axis = np.mean(real_data[self.__xposi:self.__xposi+self.__xwind,:], axis = 0)
self.x_max = np.max(self.x_axis)
self.x_min = np.min(self.x_axis)
self.mu = self.x_axis.argmax()
self.parsx.add('mu', value=self.mu)
self.parsx.add('A', value=self.x_max - self.x_min)
self.parsx.add('c', value=self.x_min)
self.parsx.add('sigma', value=self.sigma_try)
self.debug_stream('parameters fitting x axis data')
self.out_gaussx = mod.fit(self.x_axis, self.parsx, x=self.x2)
self.debug_stream('fitting x axis done succesfully')
self.a = self.out_gaussx.best_values['sigma']
return self.out_gaussx.best_values['mu']
def sinefit(self, dataIn: DataDictBase = None):
if dataIn is None:
return None
# this is just a ghetto example: only support very simple datasets
naxes = len(dataIn.axes())
ndeps = len(dataIn.dependents())
if not (naxes == 1 and ndeps == 1):
return dict(dataOut=dataIn)
# getting the data
axname = dataIn.axes()[0]
x = dataIn.data_vals(axname)
y = dataIn.data_vals(dataIn.dependents()[0])
# try to fit
sinemodel = lmfit.Model(sinefunc)
p0 = sinemodel.make_params(amp=1, freq=self.frequencyGuess, phase=0)
result = sinemodel.fit(y, p0, x=x)
# if the fit works, add the fit result to the output
dataOut = dataIn.copy()
if result.success:
dataOut['fit'] = dict(values=result.best_fit, axes=[
axname,
])
dataOut.add_meta('info', result.fit_report())
else:
dataOut.add_meta('info', 'Could not fit sine.')
return dict(dataOut=dataOut)
def fit_to_pow(self):
model = lmf.Model(self.decay_pow)
params = model.make_params(amp = 1.0, delta = 0.5)
params.add('c', value = 0, vary = False)
res_pow = model.fit(self.Fs, params, x = self.Ls)
mod = lmf.models.LinearModel()
pars = mod.guess(np.log(self.Fs), x=np.log(self.Ls))
scaling_fit = mod.fit(np.log(self.Fs),pars, x=np.log(self.Ls))
fig, ax = plt.subplots(figsize = (6, 3.6))
ax.plot(self.Ls, self.Fs, "o")
ax.plot(self.Ls, np.exp(scaling_fit.values["intercept"])*self.Ls**scaling_fit.values["slope"],"C1-")
ax.annotate("$\propto {}^{{{:.2f}}}$".format(self.L_label,scaling_fit.values["slope"]), (self.Ls[2]/2+self.Ls[3]/2, self.Fs[2]/2 + self.Fs[3]/2))
ax.set_xlabel("{}".format(self.L_label))
ax.set_ylabel("{}".format(self.F_label))
ax.locator_params(nbins=5)
ax.tick_params(axis = "y", labelleft = False)
axin = ax.inset_axes([0.4,0.4,0.6,0.6])
scaling_fit.plot_fit(ax = axin, xlabel = "log({})".format(self.L_label), ylabel = "log({})".format(self.F_label))
axin.set_title("")
axin.set_xticklabels([])
axin.set_yticklabels([])
axin.tick_params(axis='both', which='both', length=0)
axin.get_legend().remove()
plt.show()
def test_OptimizerCallback(self):
def rosen(params, a=1, b=100, noise=0):
""" Rosenbrock function """
v = (a - params[0])**2 + b * (params[1] - params[0]**2)**2
v += noise * (np.random.rand() - .5)
return v
def objective(x):
return rosen(x, 0.01, 1)
oc = OptimizerCallback(show_progress=False)
result = scipy.optimize.minimize(objective, [.5, .9],
callback=oc.scipy_callback)
self.assertEqual(result.success, True)
self.assertEqual(oc.number_of_evaluations(), result.nit)
self.assertIsInstance(oc.data, pandas.DataFrame)
oc.clear()
lmfit_model = lmfit.Model(linear_function, independent_vars=['x'])
xdata = np.linspace(0, 10, 40)
data = 10 + 5 * xdata
lmfit_model.fit(data, x=xdata, a=1, b=1, iter_cb=oc.lmfit_callback)
plt.figure(100)
plt.clf()
oc.plot(logy=True)
plt.close(100)
def fit_to_expstr(self):
model = lmf.Model(self.decay_expstr)
params = model.make_params(amp = 1.0)
params.add('c', value = 0, vary = False)
params.add('delta', value = 0.5, min = 0.1)
res_expstr = model.fit(self.Fs, params, x = self.Ls)
mod = lmf.models.LinearModel()
pars = mod.guess(np.log(self.Fs), x=np.sqrt(self.Ls))
scaling_fit = mod.fit(np.log(self.Fs)[:] ,pars, x=np.sqrt(self.Ls[:]))
fig, ax = plt.subplots(figsize = (6, 3.6))
ax.plot(self.Ls, self.Fs, "o")
ax.plot(self.Ls, np.exp(scaling_fit.values["intercept"])*np.exp(scaling_fit.values["slope"]*np.sqrt(self.Ls)),"C1-")
ax.annotate("$\propto exp({:.2f}\sqrt{{{}}})$".format(scaling_fit.values["slope"], self.L_label), (self.Ls[2]*0.1+self.Ls[3]*0.9, self.Fs[2]*0.1 + self.Fs[3]*0.9))
ax.set_xlabel("{}".format(self.L_label))
ax.set_ylabel("{}".format(self.F_label))
ax.locator_params(nbins=5)
axin = ax.inset_axes([0.35,0.35,0.6,0.6])
scaling_fit.plot_fit(ax = axin, xlabel = "$\sqrt{{{}}}$".format(self.L_label), ylabel = "log({})".format(self.F_label))
# axin.plot(self.Ls[-3:],np.log(self.Fs[-3:]),"C0o")
axin.set_title("")
axin.set_xticklabels([])
axin.set_yticklabels([])
axin.tick_params(axis='both', which='both', length=0)
axin.get_legend().remove()
plt.show()
def prepare_fitting(self):
self.fit_dicts = OrderedDict()
states = ['0', '1', '+']
for i, state in enumerate(states):
yvals = self.proc_data_dict['yvals_{}'.format(state)]
xvals = self.proc_data_dict['xvals']
mod = lmfit.Model(fit_mods.idle_error_rate_exp_decay)
mod.guess = fit_mods.idle_err_rate_guess.__get__(
mod, mod.__class__)
# Done here explicitly so that I can overwrite a specific guess
guess_pars = mod.guess(N=xvals, data=yvals)
vary_N2 = self.options_dict.get('vary_N2', True)
if not vary_N2:
guess_pars['N2'].value = 1e21
guess_pars['N2'].vary = False
# print(guess_pars)
self.fit_dicts['fit {}'.format(states[i])] = {
'model': mod,
'fit_xvals': {
'N': xvals
},
'fit_yvals': {
'data': yvals
},
'guess_pars': guess_pars
}
