r24480 - /branches/frame_order_cleanup/target_functions/frame_order.py -- July 08, 2014 - 14:21

Author: bugman
Date: Tue Jul  8 14:21:23 2014
New Revision: 24480

URL: http://svn.gna.org/viewcvs/relax?rev=24480&view=rev
Log:
Speed up of the PCS component of the rigid frame order model.

The lanthanide to atom vectors are now being calculated outside of the 
alignment tensor and spin
loops, as well as the inverse vector lengths to the 5th power.  This 
increases the speed by a factor
of 1.216 (from 38.133 to 31.368 seconds for 23329 calls of the func_rigid() 
target function).


Modified:
    branches/frame_order_cleanup/target_functions/frame_order.py

Modified: branches/frame_order_cleanup/target_functions/frame_order.py
URL: 
http://svn.gna.org/viewcvs/relax/branches/frame_order_cleanup/target_functions/frame_order.py?rev=24480&r1=24479&r2=24480&view=diff
==============================================================================
--- branches/frame_order_cleanup/target_functions/frame_order.py        
(original)
+++ branches/frame_order_cleanup/target_functions/frame_order.py        Tue 
Jul  8 14:21:23 2014
@@ -1011,10 +1011,6 @@
         # Pre-transpose matrices for faster calculations.
         RT_ave = transpose(self.R_ave)
 
-        # Pre-calculate all the necessary vectors.
-        if self.pcs_flag:
-            self.calc_vectors(pivot=self.pivot, R_ave=self.R_ave, 
RT_ave=RT_ave)
-
         # Initial chi-squared (or SSE) value.
         chi2_sum = 0.0
 
@@ -1033,6 +1029,17 @@
 
         # PCS.
         if self.pcs_flag:
+            # Pre-calculate all the necessary vectors.
+            self.calc_vectors(pivot=self.pivot, R_ave=self.R_ave, 
RT_ave=RT_ave)
+            r_ln_atom = self.r_ln_pivot + self.r_pivot_atom
+            if min(self.full_in_ref_frame) == 0:
+                r_ln_atom_rev = self.r_ln_pivot + self.r_pivot_atom_rev
+
+            # The vector length (to the inverse 5th power).
+            length = 1.0 / norm(r_ln_atom, axis=1)**5
+            if min(self.full_in_ref_frame) == 0:
+                length_rev = 1.0 / norm(r_ln_atom, axis=1)**5
+
             # Loop over each alignment.
             for align_index in range(self.num_align):
                 # Loop over the PCSs.
@@ -1041,14 +1048,14 @@
                     if not self.missing_pcs[align_index, j]:
                         # Forwards and reverse rotations.
                         if self.full_in_ref_frame[align_index]:
-                            r_pivot_atom = self.r_pivot_atom[j]
+                            r_ln_atom_i = r_ln_atom[j]
+                            length_i = length[j]
                         else:
-                            r_pivot_atom = self.r_pivot_atom_rev[j]
+                            r_ln_atom_i = r_ln_atom_rev[j]
+                            length_i = length_rev[j]
 
                         # The PCS calculation.
-                        vect = self.r_ln_pivot[0] + r_pivot_atom
-                        length = norm(vect)
-                        self.pcs_theta[align_index, j] = 
pcs_tensor(self.pcs_const[align_index, j] / length**5, vect, 
self.A_3D[align_index])
+                        self.pcs_theta[align_index, j] = 
pcs_tensor(self.pcs_const[align_index, j] * length_i, r_ln_atom_i, 
self.A_3D[align_index])
 
                 # Calculate and sum the single alignment chi-squared value 
(for the PCS).
                 chi2_sum = chi2_sum + chi2(self.pcs[align_index], 
self.pcs_theta[align_index], self.pcs_error[align_index])