comp.lang.idl-pvwave archive: archive » Re: slow processing of my k-nearest neighour code

Home » Public Forums » archive » Re: slow processing of my k-nearest neighour code

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

Re: slow processing of my k-nearest neighour code [message #49785 is a reply to message #49784]

Mon, 14 August 2006 08:08

Karl Schultz
Messages: 341
Registered: October 1999

Senior Member

On Mon, 14 Aug 2006 10:11:30 -0400, Ben Tupper wrote:

> humphreymurray@gmail.com wrote:
>
>>> ; Calculate the squared distance for each attribute.
>>> squared = make_array(num_training_elements, num_attributes)
>>> for attrib = 0, num_attributes-1 do begin
>>> squared[*,attrib] = (testing_data[i, attrib] -
>>> training_data[*,attrib])^2
>>> endfor
>>>
>
> Hi,
>
> You might try replacing the above for inner-loop with the following
>
> squared = (testing_data - training_data)^2
>
> Since IDL is array saavy it will perform the operation element by
> element for you quite quickly (as well as make the "squared" array for you).
>
> You might be able to eliminate the outer-loop, too, but I am less sure
> of that. Take a peek at the for-loop bible at
>
> http://www.dfanning.com/tips/forloops.html
>
> Good luck,
> Ben

The above will help a great deal. But if the OP is going to move to
larger data and/or it gets really important to lower the execution time,
he may want to implement a compiled-code DLM. There's a package out there
called ANN (Approximate Nearest Neighbor) that computes the k-nearest
neighbors using either an exact calculation or a faster approximate
calculation.

I needed this once as part of a surface reconstruction project and wrote a
really narrow C DLM interface wrapper that exposed the ANN features I
needed. This isn't that hard to do, but you do need to know how to code
IDL DLM's. ANN is provided in source code form and is pretty easy to work
with. And it does the job nicely.

Karl

Report message to a moderator

[Message index]

		Re: slow processing of my k-nearest neighour code By: James Kuyper on Tue, 15 August 2006 08:14
		Re: slow processing of my k-nearest neighour code By: humphreymurray on Mon, 14 August 2006 19:17
		Re: slow processing of my k-nearest neighour code By: humphreymurray on Mon, 14 August 2006 19:07
		Re: slow processing of my k-nearest neighour code By: JD Smith on Mon, 14 August 2006 14:39
		Re: slow processing of my k-nearest neighour code By: James Kuyper on Mon, 14 August 2006 13:50
		Re: slow processing of my k-nearest neighour code By: btt on Mon, 14 August 2006 09:01
		Re: slow processing of my k-nearest neighour code By: news.verizon.net on Mon, 14 August 2006 08:30
		Re: slow processing of my k-nearest neighour code By: Karl Schultz on Mon, 14 August 2006 08:08
		Re: slow processing of my k-nearest neighour code By: btt on Mon, 14 August 2006 07:11
		Re: slow processing of my k-nearest neighour code By: humphreymurray on Mon, 14 August 2006 06:50

Previous Topic:	Help with GRID_TPS
Next Topic:	Re: SHMMAP and structures

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Wed Oct 08 17:56:50 PDT 2025

Total time taken to generate the page: 0.00252 seconds