Follow-up Comment #213, task #7807 (project relax):
The attached modified 'dot_speed2.py' script shows how the current
lib.dispersion.ns_cpmg_2site_3d module at r23968 in the disp_spin_speed branch
is ~3.5 times slower than what is possible. As with previous scripts, place
this in the base relax directory (or change sys.path) and run it as 'python
dot_speed2.py'.
The script shows that current way of calculating the dot products "dot(a*1,
a*1, a)" is much slower than the first code with "a = dot(a, a)" (15.8 seconds
verses 4.6 seconds).
It also shows, just as the previous script, that the "dot(a, a, a)" notation
does not perform the inplace operation correctly and hence the printed out
matrix is incorrect, despite being the fastest of all operations (3.8
seconds). And it also shows that the "dot(a, b, c)" notation is slower than
the "c = dot(a, b)" notation?!? Numpy is clearly doing something very wrong
there!
(file #21051, file #21052)
_______________________________________________________
Additional Item Attachment:
File name: dot_speed2.py Size:5 KB
File name: dot_speed2.log Size:3 KB
_______________________________________________________
Reply to this item at:
<http://gna.org/task/?7807>
_______________________________________________
Message sent via/by Gna!
http://gna.org/
[task #7807] Speed-up of dispersion models for Clustered analysis