Thread Subject:

Randomly subsampling from matrix :(

Hi all I hope you're having nice days. I've spent way too much time on this problem... I need your help :(

I'll try to describe my problem clearly first, and then I'll just tell you exactly what I want to do. So I need to randomly subsample from a matrix. 'randsample' would be perfect, but I need to sample from a two-dimension array, and running randsample twice would mix up indices of the pairs. I thought about 'randi' to get indices first and then sample with them, but I noticed randi can return same value, which I have to avoid.

So flat out, I need to randomly sample 150 pairs of data from 250 pairs. (and I need 20 different sample sets) With studying 20 sample sets, I'm going to estimate the behavior of the original 250 pairs.

Sounds a lot like a homework right? :) yes, shamelessly I admit, this is a homework! But at least I'm doing it very much ahead of time and spent a lot of time by myself before asking here, so plz don't hate!

"Woo Kim" <efreet84@hanmir.com> wrote in message <hruf1g$qua$1@fred.mathworks.com>...
> Hi all I hope you're having nice days. I've spent way too much time on this problem... I need your help :(
>
> I'll try to describe my problem clearly first, and then I'll just tell you exactly what I want to do. So I need to randomly subsample from a matrix. 'randsample' would be perfect, but I need to sample from a two-dimension array, and running randsample twice would mix up indices of the pairs. I thought about 'randi' to get indices first and then sample with them, but I noticed randi can return same value, which I have to avoid.
>
> So flat out, I need to randomly sample 150 pairs of data from 250 pairs. (and I need 20 different sample sets) With studying 20 sample sets, I'm going to estimate the behavior of the original 250 pairs.
>
> Sounds a lot like a homework right? :) yes, shamelessly I admit, this is a homework! But at least I'm doing it very much ahead of time and spent a lot of time by myself before asking here, so plz don't hate!

a hint:
>>shuffled_choices = randperm(250);
>>indices = shuffled_choices(1:150);
>>selected_data = matrix(indices);

You'll probably need to modify this a little.

"Woo Kim" <efreet84@hanmir.com> wrote in message
news:hruf1g$qua$1@fred.mathworks.com...
> Hi all I hope you're having nice days. I've spent way too much time on
> this problem... I need your help :(
>
> I'll try to describe my problem clearly first, and then I'll just tell you
> exactly what I want to do. So I need to randomly subsample from a matrix.
> 'randsample' would be perfect, but I need to sample from a two-dimension
> array, and running randsample twice would mix up indices of the pairs. I
> thought about 'randi' to get indices first and then sample with them, but
> I noticed randi can return same value, which I have to avoid.
>
> So flat out, I need to randomly sample 150 pairs of data from 250 pairs.
> (and I need 20 different sample sets) With studying 20 sample sets, I'm
> going to estimate the behavior of the original 250 pairs.
>
> Sounds a lot like a homework right? :) yes, shamelessly I admit, this is a
> homework! But at least I'm doing it very much ahead of time and spent a
> lot of time by myself before asking here, so plz don't hate!

I agree with Sean's suggestion that you should look at RANDPERM.

M = magic(4);
numElements = numel(M);
indicesOfAllPoints = randperm(numElements);
fiveRandomElements = M(indicesOfAllPoints(1:5))

This uses "linear indexing" -- you can search the documentation for that
term for more information. Alternately, if you want to randomly select a
subset of rows:

M = magic(10);
numRows = size(M, 1);
shuffledRows = randperm(numRows);
fiveRandomRows = M(shuffledRows(1:5), :)

But the procedure you describe sounds like you might be interested in this
section of the Statistics Toolbox documentation:

http://www.mathworks.com/access/helpdesk/help/toolbox/stats/f7462.html

--
Steve Lord
slord@mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ

Thank you, Sean, and Steve Lord. That was exactly the solution I was looking for!

I did look up 'randperm' before (from another answer post by Steve Lord), but at that time I didn't think I could use it for my case... I guess this is the kind of difference between gurus and a noob.

I really enjoyed reading about Bootstrap sampling and its friends. Now I have to play around and see what kind of functions I can use for 'bootfun', but I'm pretty sure this is what I will probably end up using, also for many times in the future. Thank you so much again!