def tune_params(self, l1=None, l2=None):
"""Estimate parameters l1 and l2 based on AIC"""
if l1 == None:
l1 = numpy.linspace(numpy.percentile(abs(self.data), 25),
numpy.percentile(abs(self.data), 50), 4)
if l2 == None:
l2 = numpy.linspace(.1, .5, 3)
if self.h == None:
raise Exception(
"Need to either provide Gaussian kernel radius, h, or estimate this using the fit_radius method"
)
if self.C_ == None:
self.est_S()
AIC_results = numpy.zeros((len(l1), len(l2)))
for i in range(len(l1)):
for j in range(len(l2)):
a, b, c, AIC_results[i, j] = fitSINGLE(S=self.C_,
data=self.data,
l1=l1[i],
l2=l2[j],
pen_type=self.pen_type,
parallel=self.parallel)
index = numpy.argmin(AIC_results)
self.l1 = l1[index % len(l1)]
self.l2 = l2[index % len(l2)]
def fit(self):
#C = get_kern_cov(self.data, radius=self.h)
#C_ = [None]*len(C)
#for i in range(len(C)):
#C_[i] = C[i,:,:]
if self.C_==None:
self.est_S()
self.P, a, self.iter_, self.AIC = fitSINGLE(S=self.C_, data=self.data, l1=self.l1, l2=self.l2, parallel=self.parallel, max_iter=self.max_iter, tol=self.tol)
# fit and normalise:
if self.norm:
for i in range(len(self.P)):
d = numpy.sqrt(numpy.diagonal(self.P[i,:,:]))
self.P[i,:,:] /= numpy.outer(d,d)
def tune_params(self, l1=None, l2=None):
"""Estimate parameters l1 and l2 based on AIC"""
if l1==None:
l1 = numpy.linspace(numpy.percentile(abs(self.data), 25), numpy.percentile(abs(self.data), 50), 4)
if l2==None:
l2 = numpy.linspace(.1, .5, 3)
if self.h==None:
raise Exception("Need to either provide Gaussian kernel radius, h, or estimate this using the fit_radius method")
if self.C_==None:
self.est_S()
AIC_results = numpy.zeros((len(l1),len(l2)))
for i in range(len(l1)):
for j in range(len(l2)):
a,b,c,AIC_results[i,j] = fitSINGLE(S=self.C_, data=self.data, l1=l1[i], l2=l2[j], pen_type=self.pen_type, parallel=self.parallel)
index = numpy.argmin(AIC_results)
self.l1 = l1[ index % len(l1)]
self.l2 = l2[ index % len(l2)]
def fit(self):
#C = get_kern_cov(self.data, radius=self.h)
#C_ = [None]*len(C)
#for i in range(len(C)):
#C_[i] = C[i,:,:]
if self.C_ == None:
self.est_S()
self.P, a, self.iter_, self.AIC = fitSINGLE(S=self.C_,
data=self.data,
l1=self.l1,
l2=self.l2,
pen_type=self.pen_type,
parallel=self.parallel,
Approx=self.Approx,
max_iter=self.max_iter,
tol=self.tol)
# fit and normalise:
if self.norm:
for i in range(len(self.P)):
d = numpy.sqrt(numpy.diagonal(self.P[i, :, :]))
self.P[i, :, :] /= numpy.outer(d, d)
