def LPfit(v,i,init=dict(Te=50, Vf=15, I0=None), plot=None, maxits=None):
#Takes about 2ms/it for 2500 points, and halving this only saves 10%
if maxits is None: # this trickiness makes it work, but not pythonic.
maxits = globals()['maxits']
if plot is None: plot = debug>0
Te = init['Te']
Vf = init['Vf']
I0 = init['I0']
if I0 is None:
I0 = 1.7 * np.average(np.clip(i, 0, 1000))
var = [Te, Vf, I0]
scale = np.array([Te, Te/3, I0])
fit_results = amoeba.amoeba(var, scale, error_fun, itmax = maxits,
data = dict(v=v, i=i))
Te, Vf, I0 = fit_results[0]
plotchar(v, Te, Vf, I0, linewidth=4)
residual, mits = fit_results[1:3]
if plot:
plt.title('Te = {Te:.1f}eV, Vf = {Vf:.1f}, Isat = {I0:.3g}, resid = {r:.2g} {mits} its '
.format(Te=Te, Vf=Vf, I0=I0, r=-1*float(residual), mits=mits))
plt.suptitle('{shot} {tr} {d}'
.format(tr=t_range, shot=shot_number, d=diag_name))
plt.show(block=0)
return(fit_results)
def LPfit(v,i,init=dict(Te=50, Vf=15, I0=None), plot=None, maxits=None):
#Takes about 2ms/it for 2500 points, and halving this only saves 10%
if maxits is None: # this trickiness makes it work, but not pythonic.
maxits = globals()['maxits']
if plot is None: plot = debug>0
Te = init['Te']
Vf = init['Vf']
I0 = init['I0']
if I0 is None:
I0 = 1.7 * np.average(np.clip(i, 0, 1000))
var = [Te, Vf, I0]
scale = np.array([Te, Te/3, I0])
fit_results = amoeba.amoeba(var, scale, error_fun, itmax = maxits,
data = dict(v=v, i=i))
Te, Vf, I0 = fit_results[0]
plotchar(v, Te, Vf, I0, linewidth=4)
residual, mits = fit_results[1:3]
if plot:
plt.title('Te = {Te:.1f}eV, Vf = {Vf:.1f}, Isat = {I0:.3g}, resid = {r:.2e} {mits} its '
.format(Te=Te, Vf=Vf, I0=I0, r=-residual, mits=mits))
plt.suptitle('{shot} {tr} {d}'
.format(tr=t_range, shot=shot_number, d=diag_name))
plt.show(block=0)
return(fit_results)
def fit(self, init=None, plot=None, fit_params=None, default_init=dict(Te=50, Vf=15, I0=None)):
""" Perform a curve fit operation with the given parameter and initial
value. 'None' invokes the fit_params determined at creation
of the class, and for 'init', the default_init
"""
# Takes about 2ms/it for 2500 points, and halving this only saves 10%
# if maxits is None: # this trickiness makes it work, but not pythonic.
# maxits = globals()['maxits']
if init is None:
init = default_init
else:
default_init.update(init)
init = default_init
self.trace = []
if plot is None:
plot = self.plot
# if None, get the defaults from the __init__
if fit_params is None:
if self.fit_params is not None:
fit_params = self.fit_params
else:
raise ValueError('Need fit params in LPfitter class or call to .fit')
self.actual_fparams = self.fit_params
self.actual_fparams.update(fit_params)
alg = self.actual_fparams['alg']
maxits = self.actual_fparams['maxits']
ftol = self.actual_fparams['ftol']
xtol = self.actual_fparams['xtol']
Lnorm = self.actual_fparams['Lnorm']
Te = init['Te']
Vf = init['Vf']
I0 = init['I0']
if I0 is None:
I0 = 1.7 * np.average(np.clip(self.i, 0, 1000))
var = [Te, Vf, I0]
scale = np.array([Te, Te/3, I0])
if alg == 'leastsq':
pfit, pcov, infodict, msg, ier = \
leastsq(self.residuals, var, args=(self.v, self.i), maxfev=maxits,
full_output=1, diag=scale, ftol=ftol*I0, xtol=xtol)
# residual already raised to power Lnorm/2 = so just square here
resid = np.power(np.mean(np.abs(infodict['fvec']**2)),1./Lnorm)
# this is required to scale the covariance
s_sq = (self.residuals(pfit, self.v, self.i)**2).sum()/(len(self.i)-len(init))
fit_results = [pfit, resid, infodict['nfev']]
if ier not in [1,2,3,4]:
fit_results[-1] = 9000+ier # keep it in uint16 range
if self.parent is not None and self.parent.debug>0:
if self.parent.debug>1 or ier not in [1,2,3,4]:
print(ier, msg)
elif alg == 'amoeba':
fit_results = list(amoeba.amoeba(var, scale, self.error_fun,
itmax=maxits, ftolerance=ftol*I0,
xtolerance=xtol,
data=dict(v=self.v, i=self.i)))
fit_results[1] = -fit_results[1] # change residual back to positive
s_sq = None # don't know how to get cov for amoeba
pcov = None
Te, Vf, I0 = fit_results[0]
fit_results[1] /= I0 # normalise to IO
residual, mits = fit_results[1:3]
fit_results.append(maxits)
self.pcov = pcov
self.s_sq = s_sq
if plot > 1:
plt.scatter(self.v, self.i, c='r', s=80,
alpha=min(1, 4/np.sqrt(len(self.v))))
self.plotchar(self.v, Te, Vf, I0, linewidth=4)
# provide brief info about algorithm and Lnorm in the title
plt.title('Te = {Te:.1f}eV, Vf = {Vf:.1f}, Isat = {I0:.3g}, <res> = {r:.2e} {mits} {A}{p}:its '
.format(Te=Te, Vf=Vf, I0=I0, r=residual, mits=mits, A=alg.upper()[0],p=Lnorm))
plt.show(block=0)
return(fit_results)
