How Matlab calculate randsample using randsample function

Hi

I need to understand the method of random selection (without replacement) (MATLAB 2019b); i mean behind this function how the algorithm works?

Knowing that i have used this:

y = randsample(n,k)

Answer 1

[Adam Danz]

randsample(n,k) or randsample(population,k) merely creates a random permutation of your data (1:n or population) using randperm.

So the question becomes, how does the randperm function work?

You're in luck because it's explaind in the randperm documentation page. See the "Tips" section.

https://www.mathworks.com/help/matlab/ref/randperm.html#mw_e3144db6-fcbf-4725-bd0f-2f819b23cd93

You'll see that randperm uses a uniform pseudorandom number generator that can have different internal settings and states. There are several types of random number generators. See the table:

https://www.mathworks.com/help/matlab/math/creating-and-controlling-a-random-number-stream.html#brvku_2

To determine which one you're using, run

rng

Example output

>> rng
ans = 
  struct with fields:

     Type: 'twister'      % <------
     Seed: 9999
    State: [625×1 uint32]

For example, "twister" is defined by the mt19937ar generator which is fully described in the link above and contains citations to primary literature, depending on how far down the rabbit hole you want to go. You can also google these things and find lots of background information such as this Wiki article on the Mersenne Twister refenced in the code above.

To summarize, the randsample merely resamples or subsamples your data using randperm without replacement (unless the replacement flag is set to true). The random selection within randperm is controlled by the random number generator you're using which can be determined by running rng().

Also see this general list of topics regarding Matlab random number generation.

[riad didou]

but i still don't understand what does it mean by " uses a uniform pseudorandom number generator that can have different internal settings and states" .

for example in my case i have used this to pick 1000 Anchors form 9000 samples:

rng('default')

for it = 1:5

index = randsample(1:length(X{it}), 1000);

Anchor{it} = X{it}(index,:);

Thanks in advance

[Adam Danz]

I'm not sure what I could add without repeating myself.

The 2nd link covers the different internal settings and states.

The rng function tells you what settings you're using.

Then you can go back to the 2nd link and read some more about the internal states you're using. There are footnotes in that link pointing to the primary literature behind each method.

That's the path you need to go down to understand it deeper. I'd be glad to discuss more about something specific.

I'll add this link to my answer as well; it's a general list of topics on Matlab rng.

https://www.mathworks.com/help/matlab/random-number-generation.html

[riad didou]

Thanks a lot

[Adam Danz]

No problem. I'm not sure what you're expecting to find.

If you're trying to reproduce randomized results, you just need to set the rng seed at the beginning of your function / script.

https://www.mathworks.com/help/matlab/math/generate-random-numbers-that-are-repeatable.html