def findStepWithFit(x,data,kind="stepUp",excludePoints=100,\
order=20,fitrange=50,guessStep="digital",fitKind="Erf"):
""" Look for a step in the data
Data can be a 1D array or a 2D ones, in the latter case the 'axis' index
if used as different shot index
The 'kind' keywords should be either 'stepUp' or 'stepDown'
the 'excludePoints' keyword is used to limit the search 'excludePoints'
away from the extremes
"""
if (guessStep == "digital"):
f = timeTool.standardfilter()
pos,ampl,fwhm = timeTool.applyFilter( (data,), f, kind=kind )
idx_guess = int(pos[0])
if (idx_guess<0) or (idx_guess>(len(x)-1)): idx_guess=len(x)/2
x_guess = x[ idx_guess ]
elif (guessStep == "poly"):
x_guess = findStepWithPoly(x,data,kind=kind,excludePoints=excludePoints,order=order,fitrange=fitrange)
idx_guess = np.abs(x - x_guess).argmin()
else:
x_guess = guessStep
idx_guess = np.abs(x - x_guess).argmin()
idx = slice(np.max([0,idx_guess-fitrange]),np.min([idx_guess+fitrange,len(x)]))
xfit = x[idx]; y = data[idx]
# estimate errors by high order polinomial fit
p = np.polyfit(xfit[-10:],y[-10:],4)
err = y-np.polyval(p,xfit)
err = np.std(err)
# autoguess parameters
sig = fitrange/6.
meanLeft = np.mean(y[:10])
meanRight= np.mean(y[-10:])
a = meanRight-meanLeft
b0 = meanLeft
b1 = 0.001
if (fitKind == "Erf"):
p0 = (x_guess,a,sig,b0,b1)
fitp=optimize.curve_fit(TTfuncFit,xfit,y,p0=p0,\
maxfev=10000,ftol=1e-3, sigma=err)
names = ["steppos","amp","sig","b0","b1"]
yfit = TTfuncFit(xfit,*(fitp[0]))
else:
tau = sig*2
p0 = (x_guess,a,sig,b0,b1,tau)
fitp = optimize.curve_fit(TTfuncFitExp,xfit,y,p0=p0,maxfev=10000,
ftol=1e-3,sigma=err)
names = ["steppos","amp","sig","b0","b1","tau"]
yfit = TTfuncFitExp(xfit,*(fitp[0]))
try:
parErrs = np.sqrt( np.diag( fitp[1] ) )
except:
parErrs = np.zeros(len(names))
x0 = fitp[0][0]
if x0<xfit.min(): fitp[0][0] = 0.
ParBest = {}
ParErr = {}
for i in range(len(names)):
ParBest[ names[i] ] = fitp[0][i]
ParErr[ names[i] ] = parErrs[i]
return ParBest,ParErr,xfit,yfit
def findStepWithErfFit(x,data,kind="stepUp",excludePoints=100,order=20,fitrange=50,guessMethod="digital"):
""" Look for a step in the data
Data can be a 1D array or a 2D ones, in the latter case the 'axis' index
if used as different shot index
The 'kind' keywords should be either 'stepUp' or 'stepDown'
the 'excludePoints' keyword is used to limit the search 'excludePoints'
away from the extremes
"""
if (guessMethod == "digital"):
f = timeTool.standardfilter()
pos,ampl,fwhm = timeTool.applyFilter( (data,), f, kind=kind )
x_poly = pos[0]
#print "Digital guess",x_poly
idx = int(pos)
else:
x_poly = findStepWithPoly(x,data,kind=kind,excludePoints=excludePoints,order=order,fitrange=fitrange)
#idx = ( x>(x_poly-fitrange) ) & (x<(x_poly+fitrange) )
idx = np.abs(x - x_poly).argmin()
idx = slice(idx-fitrange,idx+fitrange)
xfit = x[idx]; y = data[idx]
# estimate errors by high order polinomial fit
p = np.polyfit(xfit[-100:],y[-100:],4)
err = y-np.polyval(p,xfit)
err = np.std(err)
# autoguess parameters
sig = fitrange/6.
meanLeft = np.mean(y[:10])
meanRight= np.mean(y[-10:])
a = meanRight-meanLeft
b0 = meanLeft
b1 = 0.001
fitp=optimize.curve_fit(TTfuncFit,xfit,y,p0=(x_poly,a,sig,b0,b1),\
maxfev=10000,ftol=1e-3, sigma=err)
(x0,a,sig,b0,b1) = fitp[0]
try:
(ex0,ea,esig,eb0,eb1) = np.sqrt( np.diag( fitp[1] ) )
except:
(ex0,ea,esig,eb0,eb1) = (0,0,0,0,0)
if (x0>xfit.max()) or (x0<xfit.min()): x0 = 0.
yfit = TTfuncFit(xfit,x0,a,sig,b0,b1)
#print "Final fit", x0,ex0
return x0,a,sig,xfit,yfit,ex0,ea,esig
