def cml_refine_agls(Prj, Ori, delta):
from copy import deepcopy
from sp_utilities import amoeba
global g_n_prj
scales = [delta] * (g_n_prj + 2)
for iprj in range(g_n_prj):
# init vec_in
vec_in = [Ori[4 * iprj], Ori[4 * iprj + 1], Ori[4 * iprj + 2]]
# prepare vec_data
vec_data = [Prj, deepcopy(Ori), iprj]
# simplex
optvec, disc, niter = amoeba(vec_in,
scales,
cml_refine_agls_wrap_dev,
data=vec_data)
# assign new angles refine
Ori[4 * iprj] = (optvec[0] + 360) % 360
Ori[4 * iprj + 1] = optvec[1]
Ori[4 * iprj + 2] = optvec[2]
sxprint('refine:', iprj, 'angles:', Ori[4 * iprj:4 * iprj + 4],
'disc:', -disc)
return Ori
def fit_tanh(dres, low=0.1):
"""
dres - list produced by the fsc funcion
dres[0] - absolute frequencies
dres[1] - fsc, because it was calculated from the dataset split into halves, convert it to full using rn = 2r/(1+r)
dres[2] - number of points use to calculate fsc coeff
low cutoff of the fsc curve
return parameters of the tanh filter: freq - cutoff frequency at which filter value is 0.5, and fall_off, the 'width' of the filter
"""
def fit_tanh_func(args, data):
pass #IMPORTIMPORTIMPORT from math import pi, tanh
v = 0.0
if (data[1][0] < 0.0):
data[1][0] *= -1.0
for i in range(len(data[0])):
fsc = 2 * data[1][i] / (1.0 + data[1][i])
if args[0] == 0 or args[1] == 0: qt = 0
else:
qt = fsc - 0.5 * (math.tanh(math.pi * (data[0][i] + args[0]) /
2.0 / args[1] / args[0]) -
math.tanh(math.pi * (data[0][i] - args[0]) /
2.0 / args[1] / args[0]))
v -= qt * qt
#print args,v
return v
setzero = False
for i in range(1, len(dres[0])):
if not setzero:
if (2 * dres[1][i] / (1.0 + dres[1][i]) < low): setzero = True
if setzero: dres[1][i] = 0.0
freq = -1.0
for i in range(1, len(dres[0]) - 1):
if ((2 * dres[1][i] / (1.0 + dres[1][i])) < 0.5):
freq = dres[0][i - 1]
break
if freq < 0.0:
# the curve never falls below 0.5, most likely something's wrong; however, return reasonable values
freq = 0.4
fall_off = 0.2
return freq, fall_off
pass #IMPORTIMPORTIMPORT from sp_utilities import amoeba
args = [freq, 0.1]
scale = [0.05, 0.05]
result = sp_utilities.amoeba(args, scale, fit_tanh_func, data=dres)
"""Multiline Comment0"""
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
#MULTILINEMULTILINEMULTILINE 0
return result[0][0], result[0][1]
def fit_tanh1(dres, low=0.1):
"""
dres - list produced by the fsc funcion
dres[0] - absolute frequencies
dres[1] - fsc, to be conservative, do not use factor of 2.
dres[2] - number of points use to calculate fsc coeff
low cutoff of the fsc curve
return parameters of the tanh filter: freq - cutoff frequency at which filter value is 0.5, and fall_off, the 'width' of the filter
"""
def fit_tanh_func(args, data):
from math import pi, tanh
v = 0.0
for i in range(len(data[0])):
fsc = data[1][i]
if args[0] == 0 or args[1] == 0: qt = 0
else:
qt = fsc - 0.5 * (
tanh(pi *
(data[0][i] + args[0]) / 2.0 / args[1] / args[0]) -
tanh(pi *
(data[0][i] - args[0]) / 2.0 / args[1] / args[0]))
v -= qt * qt
#print args,v
return v
setzero = False
for i in range(1, len(dres[0])):
if not setzero:
if (dres[1][i] < low): setzero = True
if setzero: dres[1][i] = 0.0
freq = -1.0
for i in range(1, len(dres[0]) - 1):
if (dres[1][i] < 0.5):
freq = dres[0][i - 1]
break
if (freq < 0.0):
# the curve never falls below 0.5, most likely something's wrong; however, return reasonable values
freq = 0.2
fall_off = 0.2
return freq, fall_off
from sp_utilities import amoeba
args = [freq, 0.1]
scale = [0.05, 0.05]
result = amoeba(args, scale, fit_tanh_func, data=dres)
'''
args[0] = result[0][0]
args[1] = result[0][1]
#print args
from math import pi, tanh
for i in xrange(len(dres[0])):
fsc = 2*dres[1][i]/(1.0+dres[1][i])
print i, dres[0][i],fsc , 0.5*( tanh(pi*(dres[0][i]+args[0])/2.0/args[1]/args[0]) - tanh(pi*(dres[0][i]-args[0])/2.0/args[1]/args[0]) )
#qt = fsc - 0.5*( tanh(pi*(dres[0][i]+args[0])/2.0/args[1]/args[0]) - tanh(pi*(dres[0][i]-args[0])/2.0/args[1]/args[0]) )
'''
return result[0][0], result[0][1]
def fine_2D_refinement(data, br, mask, tavg, group=-1):
# IMAGES ARE SQUARES!
nx = data[0].get_xsize()
# center is in SPIDER convention
cnx = int(old_div(nx, 2)) + 1
cny = cnx
if group > -1:
nima = 0
for im in range(len(data)):
if data[im].get_attr("ref_num") == group:
nima += 1
else:
nima = len(data)
# prepare KB interpolants
kb = kbt(nx)
# load stuff for amoeba
stuff = []
stuff.insert(0, kb)
stuff.insert(1, mask)
stuff.insert(2, nima)
# stuff.insert(3,tave) # current average
# stuff.insert(4,data) # current image in the gridding format
weights = [
br
] * 3 # weights define initial bracketing, so one would have to figure how to set them correctly
for im in range(len(data)):
if group > -1:
if data[im].get_attr("ref_num") != group:
continue
# subtract current image from the average
alpha = data[im].get_attr("alpha")
sx = data[im].get_attr("sx")
sy = data[im].get_attr("sy")
mirror = data[im].get_attr("mirror")
ddata = sp_fundamentals.prepg(data[im], kb)
ddata.set_attr_dict({
"alpha": alpha,
"sx": sx,
"sy": sy,
"mirror": mirror
})
temp = sp_fundamentals.rtshgkb(ddata, alpha, sx, sy, kb)
if mirror:
temp.process_inplace("xform.mirror", {"axis": "x"})
# Subtract current image from the average
refim = EMAN2_cppwrap.Util.madn_scalar(tavg, temp,
old_div(-1.0, float(nima)))
stuff.append(refim) # curent ave-1
stuff.append(ddata) # curent image
# perform amoeba alignment
params = [alpha, sx, sy]
outparams = sp_utilities.amoeba(params, weights, crit2d, 1.0e-4,
1.0e-4, 500, stuff)
del stuff[3]
del stuff[3]
# set parameters to the header
data[im].set_attr_dict({
"alpha": outparams[0][0],
"sx": outparams[0][1],
"sy": outparams[0][2],
"mirror": mirror,
})
# update the average
temp = sp_fundamentals.rtshgkb(ddata, outparams[0][0], outparams[0][1],
outparams[0][2], kb)
if mirror:
temp.process_inplace("xform.mirror", {"axis": "x"})
# check whether the criterion actually increased
# add current image to the average
tavg = EMAN2_cppwrap.Util.madn_scalar(refim, temp,
old_div(1.0, float(nima)))