def component_fitting(self,
mode='gaussian',
nmax=10,
full=False,
minamp=None,
alpha=0.05,
allownegative=False,
verbose=False,
save=False):
'''
Fitting to phases is much more numerically stable for von mises function
'''
fitter = lambda x, y, n: u.fit_components(
x, y, mode, n, allownegative=allownegative)
if minamp is not None:
MAX = np.max(self.data) * minamp
N = self.getNbins()
# Fit first component
n = 1
nparamA = 3
fitfunc, errfunc, pfit, perr, s_sq = fitter(self.phases, self.data, n)
residsA = self.data - fitfunc(pfit, self.phases)
RSS_funcA = np.sum(residsA**2)
def doreturn(save=False):
if save:
self.data = fitfunc(pfit, self.phases)
if full:
return fitfunc(pfit, self.phases), pfit, n
return fitfunc(pfit, self.phases)
for n in range(2, nmax + 1):
if verbose:
print("Fitting component: %i" % n)
nparamB = 3 * n
fitfunc, errfunc, pfit, perr, s_sq = fitter(
self.phases, self.data, n)
residsB = self.data - fitfunc(pfit, self.phases)
RSS_funcB = np.sum(residsB**2)
if minamp is not None and np.all(residsB < MAX):
return doreturn(save=save)
# F-test
F = ((RSS_funcA - RSS_funcB) /
(nparamB - nparamA)) / (RSS_funcB / (N - nparamB - 1))
p_value = stats.f.sf(F, nparamB - nparamA, N - nparamB - 1) #cdf?
#print F,RSS_funcA,RSS_funcB,nparamA,nparamB,p_value
if p_value > alpha: # if p_value < alpha, then B is significant, so keep going
# Replace old values
return doreturn(save=save)
nparamA = nparamB
residsA = residsB
RSS_funcA = RSS_funcB
return doreturn(save=save)
'''
def component_fitting(self,mode='gaussian',nmax=10,full=False,minamp=None,alpha=0.05,allownegative=False,verbose=False,save=False):
'''
Fitting to phases is much more numerically stable for von mises function
'''
fitter = lambda x,y,n: u.fit_components(x,y,mode,n,allownegative=allownegative)
if minamp is not None:
MAX = np.max(self.data)*minamp
N = self.getNbins()
# Fit first component
n = 1
nparamA = 3
fitfunc,errfunc,pfit,perr,s_sq = fitter(self.phases,self.data,n)
residsA = self.data - fitfunc(pfit,self.phases)
RSS_funcA = np.sum(residsA**2)
def doreturn(save=False):
if save:
self.data = fitfunc(pfit,self.phases)
if full:
return fitfunc(pfit,self.phases),pfit,n
return fitfunc(pfit,self.phases)
for n in range(2,nmax+1):
if verbose:
print("Fitting component: %i"%n)
nparamB = 3*n
fitfunc,errfunc,pfit,perr,s_sq = fitter(self.phases,self.data,n)
residsB = self.data - fitfunc(pfit,self.phases)
RSS_funcB = np.sum(residsB**2)
if minamp is not None and np.all(residsB<MAX):
return doreturn(save=save)
# F-test
F = ((RSS_funcA-RSS_funcB)/(nparamB-nparamA))/(RSS_funcB/(N-nparamB-1))
p_value = stats.f.sf(F,nparamB-nparamA,N-nparamB-1) #cdf?
#print F,RSS_funcA,RSS_funcB,nparamA,nparamB,p_value
if p_value > alpha: # if p_value < alpha, then B is significant, so keep going
# Replace old values
return doreturn(save=save)
nparamA = nparamB
residsA = residsB
RSS_funcA = RSS_funcB
return doreturn(save=save)
'''
def component_fitting(self, mode='gaussian', nmax=10):
n = 1
chisqs = 10000.0
fitter = lambda x, y, n: u.fit_components(x, y, mode, n)
while True:
fitfunc, errfunc, pfit, perr, s_sq = fitter(
self.bins, self.data, n)
#print s_sq
if s_sq < chisq:
chisq = s_sq
else:
break
n += 1
if n == nmax:
break
n -= 1
if n <= 0:
n = 1
fitfunc, errfunc, pfit, perr, s_sq = fitter(self.bins, self.data, n)
return fitfunc(pfit, self.bins)