def do_svd(self):
# make data array
data = flex.double()
for set in self.stored_series:
data.extend( set.as_1d() )
cols = self.stored_series[0].size()
rows = len(self.stored_series)
data.reshape( flex.grid(rows,cols) )
from scitbx.linalg import svd
svd_obj = svd.real( data.deep_copy(),True,True )
for ii,ss in enumerate(svd_obj.sigma):
print ii,ss, ss/flex.sum(svd_obj.sigma)
from scitbx.linalg import svd
svd_obj = svd.real( self.m.deep_copy(),True,True )
m_pseudo_inverse = svd_obj.pseudo_inverse()
cols = len(self.stored_series)
rows = 3
print rows,cols, m_pseudo_inverse.focus()
deconvol = []
for ith in range(rows):
result = self.stored_series[0]*0.0
for jset in range(cols):
result += m_pseudo_inverse[ (ith,jset) ]*self.stored_series[jset]
deconvol.append( result )
write_generic_image(deconvol,'deconvol_series.data')
write_generic_image(self.stored_series, 'series.dat')
def do_all_series(self):
for ii in range(self.navg):
self.do_series()
for ii in range(len(self.stored_series)):
self.stored_series[ii]=self.stored_series[ii]/self.navg
# do an svd analyses to find out what is going on
mean = self.stored_series[0]*0.0
new_data = flex.double()
n = len(self.stored_series)
for ii in range(n):
mean = mean+self.stored_series[ii]
mean = mean/n
for ii in range(n):
new_data.extend( (self.stored_series[ii]-mean).as_1d() )
new_data.reshape( self.grid(n,mean.size() ) )
from scitbx.linalg import svd
svd_obj = svd.real( new_data.deep_copy(),True,True )
for ii,ss in enumerate(svd_obj.sigma):
print ii,ss, ss/flex.sum(svd_obj.sigma) , "MEAN EDUCED EV's"
# get accurate fractions / kinetic data
self.m = flex.double()
for jj in range(len(self.stored_series)):
print self.vals[jj], 'series vals'
for oo in self.vals[jj]:
self.m.append(oo/self.Ntot)
self.m.reshape( flex.grid(len(self.stored_series),3) )
def rank(cls, a):
""" row rank of a """
rank_revealing = svd.real(a.deep_copy(),
accumulate_u=False, accumulate_v=False)
return rank_revealing.numerical_rank(rank_revealing.sigma[0]*1e-9)
def rank(cls, a):
""" row rank of a """
rank_revealing = svd.real(a.deep_copy(),
accumulate_u=False, accumulate_v=False)
return rank_revealing.numerical_rank(rank_revealing.sigma[0]*1e-9)
def tst_shapes():
N=80
length = 70
width=10
radius = 15
from sastbx.fXS import basic_shapes
rod = basic_shapes.rod(N,length,width,1.5)
ball = basic_shapes.ball(N,radius,1.0)
sball = basic_shapes.single_ball(N,radius,1.0)
combo = rod+ball
print "Building Rod"
rod = pad_around( enlarge(rod,1,sigma=3.0,full=True), N )
print "Building Balls"
ball = pad_around( enlarge(sball,1,sigma=3.0,full=True), N )
print "Building Combo"
combo = pad_around( enlarge(combo,1,sigma=3.0,full=True), N )
write_generic_image(rod,'rod.dat')
write_generic_image(ball,'ball.dat')
write_generic_image(combo,'combo.dat')
frod = build_scat_pat(rod)
fball = build_scat_pat(ball)
fcombo = build_scat_pat(combo)
rod_lib = scat_pat_lib(frod,N=45)
ball_lib = scat_pat_lib(fball,N=45)
combo_lib = scat_pat_lib(fcombo,N=45)
#lets build c2 functiosn for each of these buggers
np,np = fcombo.focus()
int_obj = integration_object( np )
c2_fball = int_obj.build_c2_function(fball)
c2_frod = int_obj.build_c2_function(frod)
c2_fcombo = int_obj.build_c2_function(fcombo)
write_generic_image(c2_fball, "c2_ball.dat")
write_generic_image(c2_frod, "c2_rod.dat")
write_generic_image(c2_fcombo, "c2_combo.dat")
t0_s = flex.double( [4, 5, 5] )
t1_s = flex.double( [1, 11, 7] )
t2_s = flex.double( [0, 9, 0] )
t3_s = flex.double( [0, 0, 9] )
eg = experiment_gen( [combo_lib,ball_lib,rod_lib] )
mp = mixture_prep(eg, int_obj)
mp.do_all_series()
mp.do_svd()
sys.exit()
# build synthetic images
t0_c2 = c2_fcombo*t0_s[0] + c2_fball*t0_s[1] + c2_frod*t0_s[2]
t1_c2 = c2_fcombo*t1_s[0] + c2_fball*t1_s[1] + c2_frod*t1_s[2]
t2_c2 = c2_fcombo*t2_s[0] + c2_fball*t2_s[1] + c2_frod*t2_s[2]
t3_c2 = c2_fcombo*t3_s[0] + c2_fball*t3_s[1] + c2_frod*t3_s[2]
a = flex.double( [4, 5, 5,1, 11, 7,0, 9, 0,0, 0, 9] )
a.reshape( flex.grid(4,3) )
from scitbx.linalg import svd
svd_obj = svd.real( a.deep_copy(),True,True )
a_pseudo_inverse = svd_obj.pseudo_inverse()
print list( a_pseudo_inverse )
print list(svd_obj.sigma)
rows,cols = a_pseudo_inverse.focus()
print rows, cols
cal_back = []
obs_sets = [ t0_c2 , t1_c2 , t2_c2, t3_c2 ]
deconvol = []
for ith in range(rows):
result = t0_c2*0.0
for jset in range(cols):
result += a_pseudo_inverse[ (ith,jset) ]*obs_sets[jset]
deconvol.append( result )
write_generic_image(deconvol,'deconvol.data')
MM = 10
# DO T=0
all_images = []
res = c2_fball.deep_copy()*0.0
synth = c2_fcombo*t0_s[0] + c2_fball*t0_s[1] + c2_frod*t0_s[2]
write_generic_image(synth,"t0_calc.dat")
for ii in range(MM):
print "Do image ", ii+1
ri = eg.generate_random_image(t0_s,1.0)
pc2 = int_obj.build_c2_function( ri )
res += pc2
res = res/MM
t0_c2_obs = res.deep_copy()
write_generic_image(res,"t0_obs.dat")
# Do T=1
res = c2_fball.deep_copy()*0.0
synth = c2_fcombo*t1_s[0] + c2_fball*t1_s[1] + c2_frod*t1_s[2]
write_generic_image(synth,"t1_calc.dat")
for ii in range(MM):
print "Do image ", ii+1
ri = eg.generate_random_image(t1_s,1.0)
res += int_obj.build_c2_function( ri )
res = res/MM
t1_c2_obs = res.deep_copy()
write_generic_image(res,"t1_obs.dat")
# DO T=2
res = c2_fball.deep_copy()*0.0
synth = c2_fcombo*t2_s[0] + c2_fball*t2_s[1] + c2_frod*t2_s[2]
write_generic_image(synth,"t2_calc.dat")
for ii in range(MM):
print "Do image ", ii+1
ri = eg.generate_random_image(t2_s,1.0)
res += int_obj.build_c2_function( ri )
res = res/MM
t2_c2_obs = res.deep_copy()
write_generic_image(res,"t2_obs.dat")
# DO T=3
res = c2_fball.deep_copy()*0.0
synth = c2_fcombo*t3_s[0] + c2_fball*t3_s[1] + c2_frod*t3_s[2]
write_generic_image(synth,"t3_calc.dat")
for ii in range(MM):
print "Do image ", ii+1
ri = eg.generate_random_image(t3_s,1.0)
res += int_obj.build_c2_function( ri )
res = res/MM
t3_c2_obs = res.deep_copy()
write_generic_image(res,"t3_obs.dat")
cal_back = []
obs_sets = [ t0_c2_obs , t1_c2_obs , t2_c2_obs, t3_c2_obs ]
#deconvol = []
for ith in range(rows):
result = t0_c2*0.0
for jset in range(cols):
result += a_pseudo_inverse[ (ith,jset) ]*obs_sets[jset]
deconvol.append( result )
write_generic_image(deconvol,'deconvol_obs.data')