Fastest Search in Matlab

opmtrader · Post by **opmtrader** » Fri Oct 01, 2004 8:18 pm

Hey everybody. I have a number and I'm trying to see if I can find its identical match in an array. Is this the fastest way to do so in Matlab:

if sum(the_number==the_array(:,1))>0,

If the number is in there it will return a 1. Is there a faster way? This is way faster than a for loop but I am wondering if some other function is faster than this.

Please let me know. Thanks.

monkeyA · Post by **monkeyA** » Fri Oct 01, 2004 10:36 pm

that looks the fastest... although obviously there might be a better way in general it depends on the context.

however this website is an awesome reference for vectorisation code, and may help you find another way.... its well worth a look regardless

http://www-h.eng.cam.ac.uk/help/tpl/programs/Matlab/tricks.html

opmtrader · Post by **opmtrader** » Fri Oct 01, 2004 11:00 pm

Thanks MonkeyA. Good stuff. I think I may try the "Intersect" function or one of its neighbors "Unique", "Union", or "ISMember".

I wouldn't have discovered these without your help. Thanks.

monkeyA · Post by **monkeyA** » Fri Oct 01, 2004 11:39 pm

Smiley you are very welcome

opmtrader · Post by **opmtrader** » Sat Oct 02, 2004 5:24 pm

MonkeyA, I implemented what I was doing in a new way using the Intersect function and it sped my code up immensely. I'm talking something that used to take 8 hours now takes 1 hour! At 8 hours it was completely a useless process but at 1 hour the project is now viable. Thank you again.

[ Correction: 8 hours down to a 1/2 hour! This has really made my day. Does anyone know if there is an Intersect Function in Perl? ]

monkeyA · Post by **monkeyA** » Sun Oct 03, 2004 12:45 am

thats great to hear... I have always been amazed at how much i can speed up matlab when I get down to it Smiley Out of interest, what functions did you change?... would be interested in seeing a "before-slow", "after-fast" list !!

opmtrader · Post by **opmtrader** » Mon Oct 04, 2004 10:00 pm

Well I will describe it for you.

I was reading in data from a text file. This text file had different specialized headings at each line, not standardized placement throughout.

I was accumulating data into arrays based on different categories. When I encountered a particularly commented line ("Y") I had to check to see if a part of it was also previously accumulated into another array ("X"). So I was checking this line by line (times millions of lines = millions of searches through "Y" array) using the "if sum(Y== Dead :,1))>0," command

My solution was just to collect all the "X"'s an all the "Y"'s in a single run. Then at the end of the run I simply did Intersect(X,Y) to get those values which were in both arrays.

Baltazar · Post by **Baltazar** » Tue Oct 05, 2004 8:26 pm

i just played with matlab r14

i just discovered (maybe you already know that)

the profiling tool.

it seems that it could help tune the code.

another trick is to run matlab with the tag -nojvm

in that case it does not load the java virtual machine, and this improves perf.

opmtrader · Post by **opmtrader** » Tue Oct 05, 2004 10:22 pm

Baltazar, yes I know of the Profiler. Isn't that a great tool? I used that to isolate my problem.

I will have to try this -nojvm trick. Would that be something I enter at the command line? Every last bit of speed helps. Thanks for the tip!

Baltazar · Post by **Baltazar** » Wed Oct 06, 2004 7:00 am

yep it is if you use matlab on windows, use the command line matlab.exe -nojvm

(works also on unix if i'm correct)

it disable the virtual machine, so the GUI gets simpler but the code faster