def _xval_model_error(self, fit_spectra, reject_outliers, fit_lb, fit_ub,
fitter, func):
"""
Helper function for calculation of split-half cross-validation model
error and signal reliability.
"""
set1 = fit_spectra[::2]
set2 = fit_spectra[1::2]
errs = []
signal_select = []
models = []
signals = []
for this_set in [set1, set2]:
choose_transients, model, signal, params, this_idx =\
self._fit_helper(this_set, reject_outliers,
fit_lb, fit_ub, fitter)
models.append(np.nanmean(model[choose_transients], 0))
signals.append(np.nanmean(signal[choose_transients], 0))
signal_select.append(signal[choose_transients])
#Cross-validation error estimation:
model_err = np.mean(
[ut.rmse(models[0], signals[1]),
ut.rmse(models[1], signals[0])])
# Also for the signal:
signal_err = ut.rmse(np.nanmean(signal_select[0], 0),
np.nanmean(signal_select[1], 0))
# Based on the errors, choose a function. Also report errors:
return model_err, signal_err
def _xval_choose_funcs(self, fit_spectra, reject_outliers, fit_lb, fit_ub,
fitters=[ana.fit_gaussian,ana.fit_two_gaussian],
funcs = [ut.gaussian, ut.two_gaussian]):
""" Helper function used to do split-half xvalidation to select among
alternative models"""
set1 = fit_spectra[::2]
set2 = fit_spectra[1::2]
errs = []
signal_select = []
# We can loop over functions and try each one out, checking the
# error in each:
for fitter in fitters:
models = []
signals = []
for this_set in [set1, set2]:
choose_transients, model, signal, params, this_idx =\
self._fit_helper(this_set, reject_outliers,
fit_lb, fit_ub, fitter)
models.append(np.nanmean(model[choose_transients], 0))
signals.append(np.nanmean(signal[choose_transients], 0))
signal_select.append(signal[choose_transients])
#Cross-validate!
errs.append(np.mean([ut.rmse(models[0], signals[1]),
ut.rmse(models[1], signals[0])]))
# We really only need to look at the first two:
signal_err = ut.rmse(np.nanmean(signal_select[0], 0),
np.nanmean(signal_select[1], 0))
# Based on the errors, choose a function. Also report errors:
return (fitters[np.argmin(errs)], funcs[np.argmin(errs)], np.min(errs),
signal_err)
def _xval_model_error(self, fit_spectra, reject_outliers, fit_lb, fit_ub,
fitter, func):
"""
Helper function for calculation of split-half cross-validation model
error and signal reliability.
"""
set1 = fit_spectra[::2]
set2 = fit_spectra[1::2]
errs = []
signal_select = []
models = []
signals = []
for this_set in [set1, set2]:
choose_transients, model, signal, params, this_idx =\
self._fit_helper(this_set, reject_outliers,
fit_lb, fit_ub, fitter)
models.append(np.nanmean(model[choose_transients], 0))
signals.append(np.nanmean(signal[choose_transients], 0))
signal_select.append(signal[choose_transients])
#Cross-validation error estimation:
model_err = np.mean([ut.rmse(models[0], signals[1]),
ut.rmse(models[1], signals[0])])
# Also for the signal:
signal_err = ut.rmse(np.nanmean(signal_select[0], 0),
np.nanmean(signal_select[1], 0))
# Based on the errors, choose a function. Also report errors:
return model_err, signal_err
def _xval_choose_funcs(self, fit_spectra, reject_outliers, fit_lb, fit_ub,
fitters=[ana.fit_gaussian,ana.fit_two_gaussian],
funcs = [ut.gaussian, ut.two_gaussian]):
""" Helper function used to do split-half xvalidation to select among
alternative models"""
set1 = fit_spectra[::2]
set2 = fit_spectra[1::2]
errs = []
signal_select = []
# We can loop over functions and try each one out, checking the
# error in each:
for fitter in fitters:
models = []
signals = []
for this_set in [set1, set2]:
choose_transients, model, signal, params, this_idx =\
self._fit_helper(this_set, reject_outliers,
fit_lb, fit_ub, fitter)
models.append(np.nanmean(model[choose_transients], 0))
signals.append(np.nanmean(signal[choose_transients], 0))
signal_select.append(signal[choose_transients])
#Cross-validate!
errs.append(np.mean([ut.rmse(models[0], signals[1]),
ut.rmse(models[1], signals[0])]))
# We really only need to look at the first two:
signal_err = ut.rmse(np.nanmean(signal_select[0], 0),
np.nanmean(signal_select[1], 0))
# Based on the errors, choose a function. Also report errors:
return (fitters[np.argmin(errs)], funcs[np.argmin(errs)], np.min(errs),
signal_err)