def mse_qreg(self, scale,dof,lrt,base):
'''
For debugging: returns mse for particular scale and dof, given pre-filtered lrt
'''
p = st.chi2.sf(lrt/scale, dof)
logp = sp.logn(base,p)
r = sp.logn(base,self.qnulllrtsort[0:len(lrt)])-logp
mse = (r*r).mean()
return mse
def scale_dof_obj(self,scale,dof):
base=sp.exp(1) #fitted params are invariant to this logarithm base (i.e.10, or e)
nfalse=(len(self.alteqnull)-sp.sum(self.alteqnull))
imax = int(sp.ceil(self.qmax*nfalse)) #of only non zer dof component
p = st.chi2.sf(self.lrtsort[0:imax]/scale, dof)
logp = sp.logn(base,p)
r = sp.logn(base,self.qnulllrtsort[0:imax])-logp
if self.abserr:
err=sp.absolute(r).sum()
else:#mean square error
err = (r*r).mean()
return err,imax
def symmetric_central_dense_grid(central_length, total_length, min_ds, max_ds,
max_ratio):
n_center = int(central_length / min_ds)
n_float = sc.logn(max_ratio, max_ds / min_ds)
n_geo = int(n_float + 1)
ratio = (max_ds / min_ds)**(1. / n_geo)
inc_geo_ds = np.array([min_ds * ratio**idx for idx in xrange(n_geo)])
dec_geo_ds = inc_geo_ds[::-1]
geo_length = min_ds * (ratio**n_geo - 1) / (ratio - 1)
res_length = 0.5 * (total_length - central_length) - geo_length
n_res = int(res_length / max_ds + 1)
n_total = n_center + 2 * (n_geo + n_res)
grid = np.ones(n_total, dtype=np.float64) * min_ds
indices = np.array([0, n_res, n_geo, n_center, n_geo, n_res]).cumsum()
grid[indices[0]:indices[1]] = max_ds
grid[indices[1]:indices[2]] = dec_geo_ds
grid[indices[2]:indices[3]] = min_ds
grid[indices[3]:indices[4]] = inc_geo_ds
grid[indices[4]:indices[5]] = max_ds
return grid
def to_real_page_ranks(vector):
"""Scales the items in the vector to page-rank log-scale from 1 to 10"""
minimum = min(vector)
maximum = max(vector)
span = maximum - minimum
scale_span = pow(5, 9) # the importance difference between PR 1 and PR 10
return [1 + sp.logn(5, 1 + scale_span*(float(v) - minimum)/span) for v in vector]
def gettests(L, P, C):
tests = scipy.logn(C,float(P)/L)
num = scipy.log2(tests)
if num%1.>0.:
num +=1
if num<0:
num = 0.
return long(num)
def define_bins(self, **kwargs):
r"""
This defines the bins for a logscaled histogram
"""
self.data_vector.sort()
sf = self.args['scale_fact']
num_bins = int(sp.logn(sf, self.data_vector[-1]) + 1)
#
# generating initial bins from 1 - sf**num_bins
low = list(sp.logspace(0, num_bins, num_bins + 1, base=sf))[:-1]
high = list(sp.logspace(0, num_bins, num_bins + 1, base=sf))[1:]
#
# Adding "catch all" bins for anything between 0 - 1 and less than 0
if self.data_vector[0] < 1.0:
low.insert(0, 0.0)
high.insert(0, 1.0)
if self.data_vector[0] < 0.0:
low.insert(0, self.data_vector[0])
high.insert(0, 0.0)
#
self.bins = [bin_ for bin_ in zip(low, high)]
def define_bins(self, **kwargs):
r"""
This defines the bins for a logscaled histogram
"""
self.data_vector.sort()
sf = self.args['scale_fact']
num_bins = int(sp.logn(sf, self.data_vector[-1]) + 1)
#
# generating initial bins from 1 - sf**num_bins
low = list(sp.logspace(0, num_bins, num_bins + 1, base=sf))[:-1]
high = list(sp.logspace(0, num_bins, num_bins + 1, base=sf))[1:]
#
# Adding "catch all" bins for anything between 0 - 1 and less than 0
if self.data_vector[0] < 1.0:
low.insert(0, 0.0)
high.insert(0, 1.0)
if self.data_vector[0] < 0.0:
low.insert(0, self.data_vector[0])
high.insert(0, 0.0)
#
self.bins = [bin_ for bin_ in zip(low, high)]