def Log(self):
result = BootStrap(self.nboot, name='log(' + str(self) + ')')
result.cfgvals = logNA(self.cfgvals)
result.bootvals = logNA(self.bootvals)
if hasattr(self, 'n_block') and self.n_block > 1:
result.block_cfgvals = {}
result.block_bootvals = {}
for ikey, idata in self.block_cfgvals.items():
result.block_cfgvals = logNA(self.cfgvals)
for ikey, idata in self.block_cfgvals.items():
result.block_bootvals = logNA(self.bootvals)
return result
def ChitFitFunDer(x, p):
Denom = (x[0] * (logNA(x[0]) + p[2]))**(-1)
return np.array([Denom * x[0], Denom, -x[0] * (Denom**2)])
def uWerrMine(self,data='PreDef',fun='PreDef',
Sparam='PreDef',WipeData=True,save_covar=True):
'''
Main function for computing the total autocorrelation of quantity,
following the method in [1]_.
Parameters
----------
data: {'PreDef'} [variable, (replicas), montecarlo time] optional
data to perform autocorrelation analysis over. Not passing in
any data will default to the data stored in the object
fun: {'PreDef'} function(*variable) optional
function descibing how to combine the indipendant statistical quantities.
Not passing in any function will default to the data stored in the object
Sparam: {'PreDef'} float optional
S parameter for Autocorrlation analysis
WipeData: {True} optional
Wipe the internal configuration data after computing the autocorrelation
statistical quantities.
save_covar: {True} optional
Saves the covariance matrix as an internal variable
References
----------
.. [1] U. Wolff [ALPHA Collaboration],
``Monte Carlo errors with less errors,''
Comput.\ Phys.\ Commun.\ {\bf 156}, 143 (2004)
Erratum: [Comput.\ Phys.\ Commun.\ {\bf 176}, 383 (2007)]
doi:10.1016/S0010-4655(03)00467-3, 10.1016/j.cpc.2006.12.001
[hep-lat/0306017]. '''
self.GetFuns()
if data == 'PreDef':
if isinstance(self.values,str):
print('data not imported for ',self.name)
else:
data = self.values
if fun == 'PreDef':
if isinstance(self.Fun,str):
print('Function and derivative have not been defined for ',self.name)
else:
fun,funder = self.Fun,self.FunDer
if Sparam=='PreDef': Sparam = self.Sparam
avgdata = [data[icol].mean() for icol in data]
glen = 0
if isinstance(data.index,pa.MultiIndex):
if self.has_reps:
data_numpy = []
for icol in data:
data_numpy.append([])
for istream,stream_data in data[icol].groupby(level='stream'):
# glen = max(glen,stream_data.size)
glen += stream_data.size
data_numpy[-1].append(stream_data[istream].values)
data_numpy = np.array(data_numpy)
else:
data_numpy = np.swapaxes(data.values,0,1)
glen = data_numpy.shape[1]
else:
data_numpy = np.swapaxes(data.values,0,1)
glen = data_numpy.shape[1]
## (31) matrix of autocorrelations w.r.t ab= variables
G_ab_t = []
for adat in data_numpy:
G_ab_t.append([ self.autocorr(adat,bdat) for bdat in data_numpy])
# G_ab_t.append([ self.MyCorrelate(adat,bdat) for bdat in data_numpy])
G_ab_t = np.array(G_ab_t)
if save_covar:
self.covar = []
G_ab_0 = G_ab_t[:,:,0]
for ia,Ga in enumerate(G_ab_t[:,:,0]):
for ib,Gab in enumerate(Ga):
self.covar.append(Gab/np.sqrt(G_ab_0[ia,ia]*G_ab_0[ib,ib]))
self.covar = np.array(self.covar).reshape(G_ab_0.shape)
self.covar_NoNorm = G_ab_0/glen
## (33) alpha function derivates (w.r.t variables)
f_a = np.array(funder(*avgdata))
f_ab = []
for if_a in f_a:
f_ab.append([if_b*if_a for if_b in f_a])
f_ab = np.array(f_ab)
# print 'debug' , data.shape, G_ab_t.shape, f_ab.shape, (f_ab * G_ab_t[:,:,0]).shape,np.sum(f_ab * G_ab_t[:,:,0]).shape
# print 'debug',self.name,
# print f_ab * G_ab_t[:,:,5]
# print
## (33)
GFt = [np.sum(f_ab * G_ab) for G_ab in np.rollaxis(G_ab_t,-1) ]
## (35)
CFW = np.array([GFt[0]]+[GFt[0] + 2*np.sum(GFt[1:W+1]) for W in range(1,len(GFt))])
# Bias corrections (49)
CFW = self.BiasCorrect(CFW,glen)
## (34)
nuF = CFW[0]
## equation (41)
tauint = CFW / (2*nuF)
## From Paper
tau = []
for it,itauint in enumerate(tauint):
if itauint <= 0.5:
tau.append(0.00000001)
else:
## (51)
tau.append(Sparam/logNA((2*itauint+1)/(2*itauint-1)))
# ## From Matlab
# tau = []
# for iGFt in GFt:
# if iGFt <= 0.0:
# tau.append(0.0000001)
# else:
# tau.append(Sparam/logNA((iGFt+1)/iGFt))
## (52)
Wopt = self.gW(np.array(tau),glen)
## (42)
dtauint = self.VarTau(tauint,glen)
## average, error(W), tau(W), tauerr(W), GFt(W), Wopt
## average, error[Woptimal], tau[Woptimal], tauerr[Woptimal]
self.Avg = fun(*avgdata)
self.Avg_paras = avgdata
# self.Avg = np.mean([fun(*ival) for ival in np.swapaxes(data,0,1)])
try:
self.Std = np.sqrt(np.abs(CFW[Wopt])/float(glen))
self.Std_corr = np.sqrt(np.abs(CFW[0])/float(glen))
self.Wopt = Wopt
self.GFt = GFt
self.tau = tauint[Wopt]
self.tauerr = dtauint[Wopt]
except Exception as err:
print('NaN for Autocorr')
Wopt = -1
self.Std = np.sqrt(np.abs(CFW[Wopt])/float(glen))
self.Std_corr = np.sqrt(np.abs(CFW[0])/float(glen))
self.Wopt = Wopt
self.GFt = GFt
self.tau = tauint[Wopt]
self.tauerr = dtauint[Wopt]
self.StdW = np.sqrt(np.abs(CFW)/float(glen))
self.tauW = tauint
self.tauerrW = dtauint
if WipeData:
self.RemoveVals()
self.RemoveFuns()
def ChitFitFun(x, p):
Denom = (x[0] * (logNA(x[0]) + p[2]))**(-1)
return (p[0] * x[0] + p[1]) * Denom
def LogFFDer_x0(x, p):
return [logNA(x[0])]
def LogFitFun_x0(x, p):
return p[0] * logNA(x[0])
def LogFFDer(x, p):
return [logNA(x[0]), makexunit(x[0])]
def LogFitFun(x, p):
return p[0] * logNA(x[0]) + p[1]