rng usability missing seed functionality

Folks,

It looks like using "seeds" is being considered "old fashioned" in Matlab 2022b, and there is a warning about it suggesting it will be obsolete soon. We are encouraged to use "rng" function instead, which has only one way to set the seed by using the "default" option. The problem I have with this is that my usage model is such that I want to change seeds and lock them for subsequent runs within the same program instance. I don't see how this is possibly unless I am able to save more than one "default" position. Is there another way to achieve my desired result using a function not targetted for obsolesence?

Thanks,

Kris

[Rik]

What do you mean by locking? You can specify a seed yourself if you want, the example in the documentation for rng shows you how you can do that. You seem to be confused by the syntax rng('default'). It will set the seed to 0 and the generator to twister, but you can set other parameters if you wish.

[Tahariet Sharon]

I am getting this error when using:

>> rng('default')
Error using rand
State must be a double scalar or the output of RAND('twister').

[Rik]

@Tahariet Sharon which Matlab release are you using?

[Tahariet Sharon]

2019b

[Stephen23]

@Tahariet Sharon: please show us the output of this command:

which rng -all

[Tahariet Sharon]

/Applications/MATLAB_R2022b.app/toolbox/matlab/randfun/rng.m

[Rik]

That is extremely strange. You're using R2019b, but the which command shows that the files used belongs to R2022b.

Answer 1

[Steven Lord]

It looks like using "seeds" is being considered "old fashioned" in Matlab 2022b, and there is a warning about it suggesting it will be obsolete soon.

Are you referring to this documentation page which has existed for many years now (IIRC since the introduction of rng in release R2011a over ten years ago?) That page describes why those syntaxes are discouraged, and it's not because they "old fashioned" but because they're probably not doing what you think they are.

We are encouraged to use "rng" function instead, which has only one way to set the seed by using the "default" option.

That is incorrect. You can specify a seed and a generator in rng. In fact, with your described usage you probably don't want to specify 'default', you want to explicitly fix both the seed and generator.

rng(1234, 'twister')
x = rand(1)

x = 0.1915

rng(5678, 'threefry')
y = rand(1)

y = 0.1174

rng(1234, 'twister')
z = rand(1)

z = 0.1915

x == z % true

ans = logical
   1

[Kristoffer Walker]

Thanks for pointing this out. I have been using "help rand" and "help rng" in Matlab to review the features. That is where the "default" option is mentioned, but the usage of the seeds is not as well documented there as in the online version of the help page. Just to be clear, one also uses "seed = rng" to get the current seed, and "rng('shuffle')" to reset it to a random value.