def filter_frames(self, data):
t1 = time.time()
data = data[0]
axis = self.axis
positions = self.positions
print positions
weights = data[self.peakindex]
widths = np.ones_like(positions)*self.parameters["width_guess"]
p = []
p.extend(weights)
p.extend(widths)
curvetype = self.getFitFunction(str(self.parameters['peak_shape']))
print "HI"
[x1, infodict1] = ral_nlls.solve(
p, self._resid, self.dfunc,
params=(curvetype, data, axis, positions),
options = {
'print_level': 3,
'maxit': 100,
'model': 1,
'nlls_method': 4,
'stop_g_absolute': 1e-7,
'stop_g_relative': 1e-7,
'relative_tr_radius': 0,
'initial_radius_scale': 1.0,
'maximum_radius': 1e8,
'eta_successful':1e-8,
'eta_success_but_reduce': 1e-8,
'eta_very_successful':0.9,
'eta_too_successful': 2.0,
'radius_increase': 2.0,
'radius_reduce': 0.5,
'radius_reduce_max': 1/16,
'tr_update_strategy': 2,
'hybrid_switch': 1e-0,
'scale': 0,
'scale_max':1e11,
'scale_min':1e-11,
'more_sorensen_maxits': 500,
'more_sorensen_shift': 1e-13,
'more_sorensen_tol': 1e-3,
'hybrid_tol': 2.0,
'hybrid_switch_its': 1
})
logging.debug("done one")
params = x1[0]
if np.isnan(params).any():
logging.debug('Nans were detected here')
params = np.zeros(len(params))
weights, widths, areas = self.getAreas(curvetype,
axis, positions, params)
residuals = self._resid(params, curvetype, data, axis, positions)
# all fitting routines will output the same format.
# nchannels long, with 3 elements. Each can be a subarray.
t2 = time.time()
logging.debug("Simple fit iteration took: %s ms", str((t2-t1)*1e3))
return [weights, widths, areas, residuals]
def filter_frames(self, data):
t1 = time.time()
data = data[0]
axis = self.axis
positions = self.positions
print positions
weights = data[self.peakindex]
widths = np.ones_like(positions)*self.parameters["width_guess"]
p = []
p.extend(weights)
p.extend(widths)
curvetype = self.getFitFunction(str(self.parameters['peak_shape']))
print "HI"
[x1, infodict1] = ral_nlls.solve(
p, self._resid, self.dfunc,
params=(curvetype, data, axis, positions),
options = {
'print_level': 3,
'maxit': 100,
'model': 1,
'nlls_method': 4,
'stop_g_absolute': 1e-7,
'stop_g_relative': 1e-7,
'relative_tr_radius': 0,
'initial_radius_scale': 1.0,
'maximum_radius': 1e8,
'eta_successful':1e-8,
'eta_success_but_reduce': 1e-8,
'eta_very_successful':0.9,
'eta_too_successful': 2.0,
'radius_increase': 2.0,
'radius_reduce': 0.5,
'radius_reduce_max': 1/16,
'tr_update_strategy': 2,
'hybrid_switch': 1e-0,
'scale': 0,
'scale_max':1e11,
'scale_min':1e-11,
'more_sorensen_maxits': 500,
'more_sorensen_shift': 1e-13,
'more_sorensen_tol': 1e-3,
'hybrid_tol': 2.0,
'hybrid_switch_its': 1
})
logging.debug("done one")
params = x1[0]
if np.isnan(params).any():
logging.debug('Nans were detected here')
params = np.zeros(len(params))
weights, widths, areas = self.getAreas(curvetype,
axis, positions, params)
residuals = self._resid(params, curvetype, data, axis, positions)
# all fitting routines will output the same format.
# nchannels long, with 3 elements. Each can be a subarray.
t2 = time.time()
logging.debug("Simple fit iteration took: %s ms", str((t2-t1)*1e3))
return [weights, widths, areas, residuals]
def fit(self):
"""
Run problem with RALFit.
"""
self._popt = ral_nlls.solve(self.initial_params,
self.cost_func.eval_r,
self.jacobian.eval,
options=self._options)[0]
self._status = 0 if self._popt is not None else 1
def fit(self):
"""
Run problem with RALFit.
"""
self.success = False
self._popt = ral_nlls.solve(self.initial_params,
self._prediction_error,
self._jac,
options=self._options)[0]
self.success = (self._popt is not None)
# Where H_i = [ 0 t_i x_1 e^(x_2 t_i) ]
# [ t_i x_1 e^(x_2 t_i) x_1 t_i^2 e^(x_2 t_i) ]
def Hr(x, r, t, y):
x1 = x[0]
x2 = x[1]
Hr = numpy.zeros((2, 2))
Hr[0, 0] = 0.0 # H_11
v = t * numpy.exp(x2 * t)
Hr[1, 0] = numpy.dot(r, v) # H_21
Hr[1, 1] = numpy.dot(r, (t * x1) * v) # H_22
return Hr
# Data to be fitted
t = numpy.array([1.0, 2.0, 4.0, 5.0, 8.0])
y = numpy.array([3.0, 4.0, 6.0, 11.0, 20.0])
# Starting guess
x0 = numpy.array([2.5, 0.25])
# Call fitting routine
(x, inform) = ral_nlls.solve(x0, r, J, Hr=Hr, params=(t, y))
# Print result
print("Found a local optimum at x = [ {0:.8f} {1:.8f} ]".format(x[0], x[1]))
print("RALFit converged in {:d} iterations".format(inform["iter"]))
print("The cost function is {:f} at the minimum, and ||J^Tr||={:e}".format(
inform["obj"], inform["norm_g"]))
