comp.lang.idl-pvwave archive: archive » Re: Why is MEAN so slow?

Home » Public Forums » archive » Re: Why is MEAN so slow?

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Re: Why is MEAN so slow? [message #74497 is a reply to message #74480]

Tue, 18 January 2011 18:14

Matt Francis
Messages: 94
Registered: May 2010

Member

Interesting. I've only tested this for 1 dimensional arrays on IDL
7.1, not for matrices (the application that I was trying to speed up
only used MEAN on 1D arrays).

Here is how I am comparing the two with a test code:
-----------------------
pro test_mean_speed

; Test whether the inbuilt MEAN function is slower than MEAN_QUICK

n = 100000

data = randomu(seed,100000)

indx = where(data GT 0.9,count)

data[indx] = !values.f_nan

for i=0,10000 do begin
;foo=mean_quick(data,/nan)
foo=mean(data,/nan)
endfor

end
-------------------------------

I ran this once using MEAN and once using MEAN_QUICK. Here are the
full results from PROFILER

For MEAN:

Module Type Count Only(s) Avg.(s) Time(s) Avg.
(s)
ABS (S) 10001 5.040137 0.000504 5.040137
0.000504
ARG_PRESENT (S) 10001 0.005722 0.000001 0.005722
0.000001
FINITE (S) 10001 3.328896 0.000333 3.328896
0.000333
KEYWORD_SET (S) 30003 0.014247 0.000000 0.014247
0.000000
MEAN (U) 10001 0.025580 0.000003 61.421387
0.006142
MOMENT (U) 20002 32.259783 0.001613 108.917069
0.005445
N_ELEMENTS (S) 20002 0.008905 0.000000 0.008905
0.000000
ON_ERROR (S) 30003 0.012295 0.000000 0.012295
0.000000
RANDOMU (S) 1 0.001750 0.001750 0.001750
0.001750
SIZE (S) 10001 0.011161 0.000001 0.011161
0.000001
SQRT (S) 10001 0.007367 0.000001 0.007367
0.000001
TEST_MEAN_SPEED (U) 1 0.006986 0.006986 61.430769
61.430769
TOTAL (S) 60006 18.553728 0.000309 18.553728
0.000309
WHERE (S) 10002 2.154211 0.000215 2.154211
0.000215

For MEAN_QUICK:

Module Type Count Only(s) Avg.(s) Time(s) Avg.
(s)
FINITE (S) 10001 3.247247 0.000325 3.247247
0.000325
KEYWORD_SET (S) 10001 0.004842 0.000000 0.004842
0.000000
MEAN_QUICK (U) 10001 1.432719 0.000143 9.902029
0.000990
RANDOMU (S) 1 0.001732 0.001732 0.001732
0.001732
TEST_MEAN_SPEED (U) 1 0.011449 0.011449 9.915830
9.915830
TOTAL (S) 10001 3.133684 0.000313 3.133684
0.000313
WHERE (S) 10002 2.084156 0.000208 2.084156
0.000208

According to PROFILER (as well as the obvious difference in how long
they ran for), MEAN_QUICK is much faster, for this specific problem.

Report message to a moderator

[Message index]

		Re: Why is MEAN so slow? By: Matt Francis on Wed, 19 January 2011 14:06
		Re: Why is MEAN so slow? By: Foldy Lajos on Wed, 19 January 2011 11:49
		Re: Why is MEAN so slow? By: wlandsman on Wed, 19 January 2011 11:08
		Re: Why is MEAN so slow? By: Matt Francis on Tue, 18 January 2011 18:14
		Re: Why is MEAN so slow? By: wlandsman on Tue, 18 January 2011 17:36

Previous Topic:	kriging
Next Topic:	Re: Resizeable Graphics Windows for Traditional Commands

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Fri Oct 10 20:27:47 PDT 2025

Total time taken to generate the page: 0.40130 seconds