forked from IPKM/nmreval
207-noncomplex-fits (#244)
Co-authored-by: Dominik Demuth <dominik.demuth@physik.tu-darmstadt.de> Reviewed-on: IPKM/nmreval#244 closes #207
This commit is contained in:
parent
8d3ab75c97
commit
80d9c7098c
@ -503,11 +503,25 @@ class UpperManagement(QtCore.QObject):
|
||||
we = we_option
|
||||
|
||||
if m_complex is None or m_complex == 1:
|
||||
# model is not complex: m_complex = None
|
||||
# model is complex, fit real part: m_complex = 1
|
||||
_y = data_i.y.real
|
||||
elif m_complex == 2 and np.iscomplexobj(data_i.y):
|
||||
data_complex = 1
|
||||
elif m_complex == 2:
|
||||
# model is complex, fit imag part: m_complex = 2
|
||||
if np.iscomplexobj(data_i.y):
|
||||
# data is complex, use imag part
|
||||
_y = data_i.y.imag
|
||||
data_complex = 2
|
||||
else:
|
||||
# data is real
|
||||
_y = data_i.y
|
||||
data_complex = 1
|
||||
else:
|
||||
# model is complex, fit complex: m_complex = 0
|
||||
# use data as given (complex or not)
|
||||
_y = data_i.y
|
||||
data_complex = 0
|
||||
|
||||
_x = data_i.x
|
||||
|
||||
@ -524,9 +538,9 @@ class UpperManagement(QtCore.QObject):
|
||||
|
||||
try:
|
||||
if isinstance(we, str):
|
||||
d = fit_d.Data(_x[inside], _y[inside], we=we, idx=set_id)
|
||||
d = fit_d.Data(_x[inside], _y[inside], we=we, idx=set_id, complex_type=data_complex)
|
||||
else:
|
||||
d = fit_d.Data(_x[inside], _y[inside], we=we[inside], idx=set_id)
|
||||
d = fit_d.Data(_x[inside], _y[inside], we=we[inside], idx=set_id, complex_type=data_complex)
|
||||
except Exception as e:
|
||||
raise Exception(f'Setting data failed for {set_id}')
|
||||
|
||||
|
||||
@ -6,8 +6,8 @@ from .model import Model
|
||||
from .parameter import Parameters, Parameter
|
||||
|
||||
|
||||
class Data(object):
|
||||
def __init__(self, x, y, we=None, idx=None):
|
||||
class Data:
|
||||
def __init__(self, x, y, we=None, idx=None, complex_type: int = 0):
|
||||
self.x = np.asarray(x)
|
||||
self.y = np.asarray(y)
|
||||
if self.y.shape[0] != self.x.shape[0]:
|
||||
@ -20,6 +20,7 @@ class Data(object):
|
||||
self.parameter = Parameters()
|
||||
self.para_keys: list = []
|
||||
self.fun_kwargs = {}
|
||||
self.complex_type = complex_type
|
||||
|
||||
def __len__(self):
|
||||
return self.y.shape[0]
|
||||
|
||||
@ -361,7 +361,7 @@ class FitRoutine(object):
|
||||
with np.errstate(all='ignore'):
|
||||
res = optimize.least_squares(cost, p0, bounds=(lb, ub), max_nfev=500 * len(p0))
|
||||
|
||||
err, corr, partial_corr = self._calc_error(res.jac, np.sum(res.fun**2), *res.jac.shape)
|
||||
err, corr, partial_corr = _calc_error(res.jac, np.sum(res.fun**2), *res.jac.shape)
|
||||
self.make_results(data, res.x, var, data.para_keys, res.jac.shape,
|
||||
err=err, corr=corr, partial_corr=partial_corr)
|
||||
|
||||
@ -375,7 +375,7 @@ class FitRoutine(object):
|
||||
with np.errstate(all='ignore'):
|
||||
res = optimize.least_squares(cost, p0, bounds=(lb, ub), max_nfev=500 * len(p0))
|
||||
|
||||
err, corr, partial_corr = self._calc_error(res.jac, np.sum(res.fun**2), *res.jac.shape)
|
||||
err, corr, partial_corr = _calc_error(res.jac, np.sum(res.fun**2), *res.jac.shape)
|
||||
for v, var_pars_k in zip(data, data_pars):
|
||||
self.make_results(v, res.x, var, var_pars_k, res.jac.shape,
|
||||
err=err, corr=corr, partial_corr=partial_corr)
|
||||
@ -458,9 +458,17 @@ class FitRoutine(object):
|
||||
self.make_results(v, res.beta, var, var_pars_k, (sum(len(d) for d in data), len(p0)),
|
||||
err=res.sd_beta, corr=corr, partial_corr=partial_corr)
|
||||
|
||||
def make_results(self, data, p, var_pars, used_pars, shape,
|
||||
err=None, corr=None, partial_corr=None):
|
||||
|
||||
def make_results(
|
||||
self,
|
||||
data: Data,
|
||||
p: list[float],
|
||||
var_pars: list[str],
|
||||
used_pars: list[str],
|
||||
shape: tuple[int, int],
|
||||
err: list[float] = None,
|
||||
corr: np.ndarray = None,
|
||||
partial_corr: np.ndarray = None,
|
||||
):
|
||||
if err is None:
|
||||
err = [0] * len(p)
|
||||
|
||||
@ -498,22 +506,24 @@ class FitRoutine(object):
|
||||
model = data.get_model()
|
||||
|
||||
self.result[idx] = FitResultCreator.make_with_model(
|
||||
model,
|
||||
data.x,
|
||||
data.y,
|
||||
actual_parameters,
|
||||
data.fun_kwargs,
|
||||
data.we_string,
|
||||
data.idx,
|
||||
*shape,
|
||||
model=model,
|
||||
x_orig=data.x,
|
||||
y_orig=data.y,
|
||||
p=actual_parameters,
|
||||
fun_kwargs=data.fun_kwargs,
|
||||
we=data.we_string,
|
||||
idx=data.idx,
|
||||
nobs=shape[0],
|
||||
nvar=shape[1],
|
||||
corr=actual_corr,
|
||||
pcorr=actual_pcorr,
|
||||
data_mode=data.complex_type,
|
||||
)
|
||||
|
||||
return self.result
|
||||
|
||||
@staticmethod
|
||||
def _calc_error(jac, chi, nobs, nvars):
|
||||
|
||||
def _calc_error(jac, chi, nobs, nvars):
|
||||
# copy of scipy.curve_fit to calculate covariance
|
||||
# noinspection PyTupleAssignmentBalance
|
||||
try:
|
||||
|
||||
@ -9,7 +9,7 @@ from ._meta import MultiModel
|
||||
from .parameter import Parameters, Parameter
|
||||
|
||||
|
||||
class Model(object):
|
||||
class Model:
|
||||
def __init__(self, model, *args, **kwargs):
|
||||
self.idx = kwargs.pop('idx', None)
|
||||
|
||||
|
||||
@ -11,6 +11,7 @@ from scipy.stats import f as fdist
|
||||
from scipy.interpolate import interp1d
|
||||
|
||||
from ._meta import MultiModel
|
||||
from .model import Model
|
||||
from .parameter import Parameter
|
||||
from ..data.points import Points
|
||||
from ..data.signals import Signal
|
||||
@ -36,17 +37,30 @@ class FitResultCreator:
|
||||
else:
|
||||
resid = kwargs['y'] - y_orig
|
||||
|
||||
stats = FitResultCreator.calc_statistics(resid, _y)
|
||||
stats = calc_statistics(resid, _y)
|
||||
|
||||
return FitResult(kwargs['x'], kwargs['y'], x_orig, y_orig, params, dict(kwargs['choice']), resid, 0, 0,
|
||||
kwargs['name'], stats, idx)
|
||||
return FitResult(
|
||||
x=kwargs['x'],
|
||||
y=kwargs['y'],
|
||||
x_data=x_orig,
|
||||
y_data=y_orig,
|
||||
params=params,
|
||||
fun_kwargs=dict(kwargs['choice']),
|
||||
resid=resid,
|
||||
nobs=0,
|
||||
nvar=0,
|
||||
we='',
|
||||
name=kwargs['name'],
|
||||
stats=stats,
|
||||
idx=idx,
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def make_with_model(
|
||||
model: 'Model',
|
||||
x_orig: np.ndarray,
|
||||
y_orig: np.ndarray,
|
||||
p: 'Parameters',
|
||||
p: list,
|
||||
fun_kwargs: dict,
|
||||
we: str,
|
||||
idx: str | None,
|
||||
@ -54,6 +68,7 @@ class FitResultCreator:
|
||||
nvar: int,
|
||||
corr: np.ndarray,
|
||||
pcorr: np.ndarray,
|
||||
data_mode: int,
|
||||
) -> FitResult:
|
||||
if np.all(x_orig > 0) and (np.max(x_orig) > 100 * np.min(x_orig)):
|
||||
islog = True
|
||||
@ -83,17 +98,11 @@ class FitResultCreator:
|
||||
actual_mode = fun_kwargs['complex_mode']
|
||||
fun_kwargs['complex_mode'] = 0
|
||||
|
||||
_y = model.func(p_final, _x, **fun_kwargs)
|
||||
|
||||
if not actual_mode < 0:
|
||||
if actual_mode == 1:
|
||||
_y.imag = 0
|
||||
elif actual_mode == 2:
|
||||
_y.real = 0
|
||||
_y = check_complex(model.func(p_final, _x, **fun_kwargs), actual_mode, data_mode)
|
||||
|
||||
fun_kwargs['complex_mode'] = actual_mode
|
||||
|
||||
stats = FitResultCreator.calc_statistics(_y, resid, nobs, nvar)
|
||||
stats = calc_statistics(_y, resid, nobs, nvar)
|
||||
varied = [p.var for p in parameters.values()]
|
||||
|
||||
if corr is None:
|
||||
@ -134,38 +143,9 @@ class FitResultCreator:
|
||||
pcorr=partial_correlation,
|
||||
islog=islog,
|
||||
func=model,
|
||||
data_complex=data_mode,
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def calc_statistics(y, residual, nobs=None, nvar=None):
|
||||
chi = (residual**2).sum()
|
||||
try:
|
||||
r = 1 - chi/((y-np.mean(y))**2).sum()
|
||||
except RuntimeWarning:
|
||||
r = -9999
|
||||
|
||||
if nobs is None:
|
||||
nobs = 1
|
||||
|
||||
if nvar is None:
|
||||
nvar = 0
|
||||
|
||||
dof = nobs - nvar
|
||||
loglikehood = nobs * np.log(chi / nobs)
|
||||
|
||||
stats = {
|
||||
'chi^2': chi,
|
||||
'R^2': r,
|
||||
'AIC': loglikehood + 2 * nvar,
|
||||
'BIC': loglikehood + np.log(nobs) * nvar,
|
||||
'adj. R^2': 1 - (nobs-1) / (dof+1e-13) * (1-r),
|
||||
'red. chi^2': chi / (dof + 1e-13),
|
||||
}
|
||||
|
||||
stats['AICc'] = stats['AIC'] + 2*(nvar+1)*nvar / (dof - 1 + 1e-13)
|
||||
|
||||
return stats
|
||||
|
||||
|
||||
class FitResult(Points):
|
||||
|
||||
@ -188,7 +168,8 @@ class FitResult(Points):
|
||||
pcorr: np.ndarray = None,
|
||||
islog: bool = False,
|
||||
func=None,
|
||||
**kwargs
|
||||
data_complex: int = 1,
|
||||
**kwargs,
|
||||
):
|
||||
|
||||
self.parameter, name = self._prepare_names(params, name)
|
||||
@ -210,6 +191,7 @@ class FitResult(Points):
|
||||
self.y_data = y_data
|
||||
self._model_name = name
|
||||
self._func = func
|
||||
self._data_complex = data_complex
|
||||
|
||||
@staticmethod
|
||||
def _prepare_names(parameter: dict, modelname: str):
|
||||
@ -418,20 +400,9 @@ class FitResult(Points):
|
||||
if self.func is None:
|
||||
raise ValueError('no fit function available to calculate new y values')
|
||||
|
||||
actual_mode = -1
|
||||
if 'complex_mode' in self.fun_kwargs:
|
||||
actual_mode = self.fun_kwargs['complex_mode']
|
||||
self.fun_kwargs['complex_mode'] = 0
|
||||
|
||||
new_fit = self.copy()
|
||||
y_values = self.func.func(self.p_final, x_values, **self.fun_kwargs)
|
||||
if not actual_mode < 0:
|
||||
if actual_mode == 1:
|
||||
y_values.imag = 0
|
||||
elif actual_mode == 2:
|
||||
y_values.real = 0
|
||||
|
||||
self.fun_kwargs['complex_mode'] = actual_mode
|
||||
y_values = check_complex(y_values, self.fun_kwargs.get('complex_mode', -1), self._data_complex)
|
||||
|
||||
new_fit.set_data(x_values, y_values, y_err=0.0)
|
||||
|
||||
@ -442,20 +413,13 @@ class FitResult(Points):
|
||||
raise ValueError('no fit function available to calculate new y values')
|
||||
|
||||
part_functions = []
|
||||
actual_mode = -1
|
||||
if 'complex_mode' in self.fun_kwargs:
|
||||
actual_mode = self.fun_kwargs['complex_mode']
|
||||
self.fun_kwargs['complex_mode'] = 0
|
||||
actual_mode = self.fun_kwargs.get('complex_mode', -1)
|
||||
|
||||
for sub_name, sub_y in zip(self.func.sub_name(), self.func.sub(self.p_final, x_values, **self.fun_kwargs)):
|
||||
if not actual_mode < 0:
|
||||
if actual_mode == 1:
|
||||
sub_y.imag = 0
|
||||
elif actual_mode == 2:
|
||||
sub_y.real = 0
|
||||
sub_y = check_complex(sub_y, actual_mode, self._data_complex)
|
||||
|
||||
if np.iscomplexobj(sub_y):
|
||||
part_functions.append(Signal(x_values, sub_y, name=sub_name))
|
||||
|
||||
else:
|
||||
part_functions.append(Points(x_values, sub_y, name=sub_name))
|
||||
|
||||
@ -463,3 +427,49 @@ class FitResult(Points):
|
||||
self.fun_kwargs['complex_mode'] = actual_mode
|
||||
|
||||
return part_functions
|
||||
|
||||
|
||||
def check_complex(y, model_complex, data_complex):
|
||||
if not np.iscomplexobj(y):
|
||||
return y
|
||||
|
||||
if model_complex == 1:
|
||||
y.imag = 0
|
||||
if data_complex == 1:
|
||||
y = y.real
|
||||
elif model_complex == 2:
|
||||
y.real = 0
|
||||
if data_complex == 1:
|
||||
y = y.imag
|
||||
|
||||
return y
|
||||
|
||||
|
||||
def calc_statistics(y, residual, nobs=None, nvar=None):
|
||||
chi = (residual**2).sum()
|
||||
try:
|
||||
r = 1 - chi/((y-np.mean(y))**2).sum()
|
||||
except RuntimeWarning:
|
||||
r = -9999
|
||||
|
||||
if nobs is None:
|
||||
nobs = 1
|
||||
|
||||
if nvar is None:
|
||||
nvar = 0
|
||||
|
||||
dof = nobs - nvar
|
||||
loglikehood = nobs * np.log(chi / nobs)
|
||||
|
||||
stats = {
|
||||
'chi^2': chi,
|
||||
'R^2': r,
|
||||
'AIC': loglikehood + 2 * nvar,
|
||||
'BIC': loglikehood + np.log(nobs) * nvar,
|
||||
'adj. R^2': 1 - (nobs-1) / (dof+1e-13) * (1-r),
|
||||
'red. chi^2': chi / (dof + 1e-13),
|
||||
}
|
||||
|
||||
stats['AICc'] = stats['AIC'] + 2*(nvar+1)*nvar / (dof - 1 + 1e-13)
|
||||
|
||||
return stats
|
||||
|
||||
Loading…
Reference in New Issue
Block a user