-
Notifications
You must be signed in to change notification settings - Fork 0
/
optimization_notes.txt
45 lines (34 loc) · 1.48 KB
/
optimization_notes.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
current optimum: only run 3D-matrix parts (Z) only of scov.m on GPU.
run_classification G preprocessed: 175 sec per participant
=================================== scov ==================================
normal cov on GPU? --> no
- CPU: 1.432 / 20.335
- GPU: 2.406 / 20.972
whole scov on GPU (e.g. half of the workers)? --> nope
one participant without GPU: 858s
with GPU: 192s (only in scov: meaned2 until nom)
===========================================================================
================================= workers =================================
4 workers: 12 sec
5 workers: 18 sec
6 workers: 21 sec
8 workers: 24 sec
4 workers a 2 threads: 20 sec
==> use 4 workers a 1 thread
===========================================================================
=================================== eig ===================================
toeig = gpuArray(within_class_scatter) \ gpuArray(between_class_scatter);
phi = gather(eig(toeig));
eig: 14.476
\: 0.758
---------------------------------------------------------------------------
toeig = within_class_scatter \ between_class_scatter;
phi = eig(toeig);
eig: 14.665
\: 0.857
---------------------------------------------------------------------------
phi = eig(between_class_scatter, within_class_scatter);
eig: 1.956
---------------------------------------------------------------------------
==> clearly: run two-argument version on CPU!
===========================================================================