def filter_pairs( h1h2, c2a, B, BI, tol = 1e-5):
order = np.argsort( c2a ) # increasing
h1, h2 = h1h2[order[0]]
BT = BTmat( h1, h2, B, BI )
UBI = np.array( (np.dot(B, h1), np.dot(B,h2), (0,0,0)), float )
cImageD11.quickorient( UBI, BT )
UB = np.linalg.inv( UBI )
pairs = [ (h1, h2), ]
cangs = [ c2a[order[0]], ]
matrs = [ BT, ]
gvecs = [ np.dot( UB, HKL0 ).T.copy(), ]
nhkl = len(HKL0[0])
#import pdb;pdb.set_trace()
for i in order[1:]:
h1, h2 = h1h2[i]
BT = BTmat( h1, h2, B, BI )
UBI = np.array( (np.dot(B, h1), np.dot(B,h2), (0,0,0)), float )
cImageD11.quickorient( UBI, BT )
j = len(pairs)-1
while j>=0 and fabs(c2a[order[i]] - cangs[j]) < 1e-6:
if cImageD11.score( UBI, gvecs[j], 1e-8 ) < nhkl+1:
pairs += [ (h1, h2), ]
cangs += [ c2a[i], ]
matrs += [ BT, ]
gvecs += [ np.dot( np.linalg.inv(UBI) , HKL0 ).T.copy(), ]
j = -1 # break
else:
j -= 1
return pairs, cangs, matrs
def orient(self, ring1, g1, ring2, g2, verbose=0, all=True):
"""
Compute an orientation matrix using cell parameters and the indexing
of two reflections
Orientation matrix:
A matrix such that h = A^1 g
define 3 vectors t1,t2,t3
t1 is parallel to first peak (unit vector along g1)
t2 is in the plane of both (unit vector along g1x(g1xg2))
t3 is perpendicular to both (unit vector along g1xg2)
"""
costheta = np.dot(g1, g2) / np.sqrt((g1 * g1).sum() * (g2 * g2).sum())
hab, c2ab, matrs = self.getanglehkls(ring1, ring2)
if all:
lo = min(np.searchsorted(c2ab, costheta - 1e-5, side='left'),
len(c2ab) - 1)
hi = min(np.searchsorted(c2ab, costheta + 1e-5, side='right'),
len(c2ab))
best = range(lo, hi)
if verbose == 1:
print("lo hi best", lo, hi, best, len(c2ab))
else:
i = np.searchsorted(c2ab, costheta, side='left')
if i > 0 and (
i == len(c2ab) or
(fabs(costheta - c2ab[i - 1]) < fabs(costheta - c2ab[i]))):
best = [
i - 1,
]
else:
best = [
i,
]
if verbose == 1:
print("g1, g2", g1, g2)
print("observed cos2theta", costheta)
print("hab, c2ab", hab, c2ab)
print("best", best)
self.UBIlist = []
UBlist = []
for b in best:
h1, h2 = hab[b]
if verbose == 1:
print("Assigning h1",h1,g1,self.ds(h1),\
math.sqrt(np.dot(g1,g1)),\
self.ds(h1)-math.sqrt(np.dot(g1,g1)))
print("Assigning h2",h2,g2,self.ds(h2),\
math.sqrt(np.dot(g2,g2)),\
self.ds(h1)-math.sqrt(np.dot(g1,g1)))
print("Cos angle calc", self.anglehkls(h1, h2), "obs",
costheta, "c2ab", c2ab[b])
BT = matrs[b]
UBI = np.empty((3, 3), float)
UBI[0] = g1
UBI[1] = g2
cImageD11.quickorient(UBI, BT)
if verbose == 1:
print("UBI")
print(UBI)
h = np.dot(UBI, g1)
print("(%9.3f, %9.3f, %9.3f)" % (h[0], h[1], h[2]))
h = np.dot(UBI, g2)
print("(%9.3f, %9.3f, %9.3f)" % (h[0], h[1], h[2]))
self.UBI = UBI
self.UB = np.linalg.inv(UBI)
self.UBIlist.append(UBI)
UBlist.append(self.UB)
# trim to uniq list? What about small distortions...
self.UBIlist = ubi_equiv(self.UBIlist, UBlist)
def filter_pairs( h1, h2, c2a, B, BI, tol = 1e-5):
""" remove duplicate pairs for orientation searches
h1 = reflections of ring1, N1 peaks
h2 = reflections of ring2, N2 peaks
c2a = cos angle between them, N1xN2
B = B matrix in reciprocal space
BI = inverse in real space
"""
assert c2a.shape == (len(h1), len(h2))
order = np.argsort( c2a.ravel() ) # increasing in cosine of angle
c2as = c2a.flat[order]
hi, hj= np.mgrid[0:len(h1),0:len(h2)]
hi = hi.ravel()[order] # to get back the peaks
hj = hj.ravel()[order]
# Results holders:
pairs = [ ]
cangs = [ ]
matrs = [ ]
# cluster the cangs assuming a sensible threshold
dc = (c2as[1:] - c2as[:-1]) > 1e-8 # differences
inds = list(np.arange( 1, len(dc) + 1, dtype=int )[dc]) + [len(c2as)-1,]
p = 0 # previous
for i in inds:
c = c2as[p:i] # block is p:i
if abs(c2as[p]) < 0.98: # always keep the first one
ha = h1[hi[p]]
hb = h2[hj[p]]
pairs.append( (ha, hb) )
cangs.append( c2as[p] )
BT = BTmat( ha, hb, B, BI)
matrs.append( BT )
else:
p = i
continue
if len(c) == 1:
p = i
continue
assert (c.max()-c.min()) < 2.1e-8, "Angles blocking error in filter_pairs"
# here we have a bunch of hkl pairs which all give the same angle
# between them. They are not all the same. We generate a pair of peaks
# from the first one and see which other pairs index differently
ga = np.dot(B, ha )
gb = np.dot(B, hb )
assert abs( np.dot(ga,gb)/np.sqrt(np.dot(ga,ga)*np.dot(gb,gb)) - c2as[p] ) < 2e-8, "mixup in filter_pairs"
gobs = np.array( (ga, gb, (0,0,0)), float)
UBI = gobs.copy()
cImageD11.quickorient( UBI, BT )
gtest = [ np.dot( np.linalg.inv(UBI), HKL0 ).T, ]
for j in range(p+1,i):
ha = h1[hi[j]]
hb = h2[hj[j]]
BT = BTmat( ha, hb, B, BI)
newpair = True
for gt in gtest:
UBI = gobs.copy()
cImageD11.quickorient( UBI, BT )
npk = cImageD11.score(UBI, gt, 1e-6)
if npk == len(HKL0[0]):
newpair = False
break
if newpair:
pairs.append( (ha, hb) )
cangs.append( c2as[j] )
matrs.append( BT )
gtest.append( np.dot( np.linalg.inv(UBI), HKL0 ).T )
p = i
return pairs, cangs, matrs