Lua random seed - random

I have a python program which I am trying to replicate in lua. In the python code, I have created a number of random matrices and have initially set the random seed using np.random.seed(seed=1). To make it easier to debug and trace the output of my lua program, I would like to set the same random seed in my lua code also. Is this possible and if so, how?
In lua, I have tried:
math.randomseed(1)
Then, I set a=torch.rand(2,2) and printed a but different matrices are being printed when I print a
What I need:
In python, when I set random seed as 1 and print a 2d matrix, I get:
([[ 4.17022005e-01, 7.20324493e-01],
[ 1.14374817e-04, 3.02332573e-01]])
When I print a 2d matrix in lua, I must get the same matrix(after using the same seed)
Is this possible?

Looks like you need to use torch.manualSeed(1) for torch random
http://torch7.readthedocs.io/en/rtd/random/

Related

Is srand() required in Julia?

Some low-level languages like C require the programmer to set seed (usually srand(time(0)) if the user wants a different sequence of random numbers whenever the program runs. If it is not set, the program generates the same sequence of random numbers for each run.
Some high-level languages automatically set the seed if it is not set at first.
In Julia, if I want to generate a new sequence of random numbers each time, should I call srand()?
If you call Julia's srand() without providing a seed, Julia will use system entropy for seeding (essentially using a random seed).
On startup (specifically during initialisation of the Random module), Julia calls srand() without arguments. This means the global RNG is initialised randomly.
That means there's usually no need to call srand() in your own code unless you want to make the point that your random results are not meant to be reproducible.
Julia seeds the random number generator automatically, you use srand with a known seed, in order to recreate the same pseudo random sequence deterministically (useful for testing for example), but if you want to generate a different random sequence each time, all you need is to call rand.
help?> srand
search: srand sprand sprandn isreadonly StepRange StepRangeLen ClusterManager AbstractRNG AbstractUnitRange CartesianRange
srand([rng=GLOBAL_RNG], seed) -> rng
srand([rng=GLOBAL_RNG]) -> rng
Reseed the random number generator: rng will give a reproducible sequence
of numbers if and only if a seed is provided. Some RNGs
don't accept a seed, like RandomDevice. After the call to srand, rng is
equivalent to a newly created object initialized with the
same seed.

In octave poissrnd ignores the seed that has been set

I am running octave 3.8.1. Even if I set the seed for the random number generator, poissrnd always produce a different number. Let us consider the following code, for example
for i=1:2
rand('state',1); randn('state',1);
poissrnd(10)
end
Running it in matlab, produce the same number in both iterations. Running it in Octave, always produce a different number.
How can I correctly set a seed to poissrnd?
Thank you
Ok, I found the solution. You have to use randp('state',1). Therefore, the script
for i=1:2
randp('state',1);
poissrnd(10)
end
would always produce the same numbers.

Rust GSL library always returns the same number for a random number generator

I am using the rgsl library in Rust that wraps functions from the C GSL math libraries. I was using a random number generator function, but I am always getting the same exact value whenever I generate a new random number. I imagine that the number should vary upon each run of the function. Is there something that I am missing? Do I need to set a new random seed each time or such?
extern crate rgsl;
use rgsl::Rng;
fn main() {
rgsl::RngType::env_setup();
let t = rgsl::rng::default();
let r = Rng::new(&t).unwrap()
let val = rgsl::randist::binomial::binomial(&r, 0.01f64, 1u32);
print!("{}",val);
}
The value I keep getting is 1, which seems really high considering the probability of obtaining a 1 is 0.01.
The documentation for env_setup explains everything you need to know:
This function reads the environment variables GSL_RNG_TYPE and GSL_RNG_SEED and uses their values to set the corresponding library variables gsl_rng_default and gsl_rng_default_seed
If you don’t specify a generator for GSL_RNG_TYPE then gsl_rng_mt19937 is used as the default. The initial value of gsl_rng_default_seed is zero.
(Emphasis mine)
Like all software random number generators, this is really an algorithm that produces pseudo random numbers. The algorithm and the initial seed uniquely identify a sequence of these numbers. Since the seed is always the same, the first (and second, third, ...) number in the sequence will always be the same.
So if I want to generate a new series of random numbers, then I need to change the seed each time. However, if I use the rng to generate a set of random seeds, then I will get the same seeds each time.
That's correct.
Other languages don't seem to have this constraint, meaning that the seed can be manually set if desired, but is otherwise is random.
A classical way to do this is to seed your RNG with the current time. This produces an "acceptable" seed for many cases. You can also get access to true random data from the operating system and use that as a seed or mix it in to produce more random data.
Is there no way to do this in Rust?
This is a very different question. If you just want a random number generator in Rust, use the rand crate. This uses techniques like I described above.
You could even do something crazy like using random values from the rand crate to seed your other random number generator. I just assumed that there is some important reason you are using that crate instead of rand.

Is it possible to reverse a pseudo random number generator?

Is it possible to reverse a pseudo random number generator?
For example, take an array of generated numbers and get the original seed.
If so, how would this be implemented?
This is absolutely possible - you just have to create a PRNG which suits your purposes. It depends on exactly what you need to accomplish - I'd be happy to offer more advice if you describe your situation in more detail.
For general background, here are some resources for inverting a Linear Congruential Generator:
Reversible pseudo-random sequence generator
pseudo random distribution which guarantees all possible permutations of value sequence - C++
And here are some for inverting the mersenne twister:
http://www.randombit.net/bitbashing/2009/07/21/inverting_mt19937_tempering.html
http://b10l.com/reversing-the-mersenne-twister-rng-temper-function/
In general, no. It should be possible for most generators if you have the full array of numbers. If you don't have all of the numbers or know which numbers you have (do you have the 12th or the 300th?), you can't figure it out at all, because you wouldn't know where to stop.
You would have to know the details of the generator. Decoding a linear congruential generator is going to be different from doing so for a counter-based PRNG, which is going to be different from the Mersenne twister, which is going to be different with a Fibonacci generator. Plus you would probably need to know the parameters of the generator. If you had all of that AND the equation to generate a number is invertible, then it is possible. As to how, it really depends on the PRNG.
Use the language Janus a time-reversible language for doing reversible computing.
You could probably do something like create a program that does this (pseudo-code):
x = seed
x = my_Janus_prng(x)
x = reversible_modulus_op(x, N) + offset
Janus has the ability to give to you a program that takes the output number and whatever other data it needs to invert everything, and give you the program that ends with x = seed.
I don't know all the details about Janus or how you could do this, but just thought I would mention it.
Clearly, what you want to do is probably a better idea because if the RNG is not an injective function, then what should it map back to etc.
So you want to write a Janus program that outputs an array. The input to the Janus inverted program would then take an array (ideally).

How to implement a pseudo random function

I want to generate a sequence of random numbers that will be used to pick tiles for a "maze". Each maze will have an id and I want to use that id as a seed to a pseudo random function. That way I can generate the same maze over and over given it's maze id. Preferably I do not want to use a built in pseudo random function in a language since I do not have control over the algorithm and it could change from platform to platform. As such, I would like to know:
How should I go about implementing my own pseudo random function?
Is it even feasible to generate platform independent pseudo random numbers?
Yes, it is possible.
Here is an example of such an algorithm (and its use) for noise generation.
Those particular random functions (Noise1, Noise2, Noise3, ..) use input parameters and calculate the pseudo random values from there.
Their output range is from 0.0 to 1.0.
And there are many more out there (Like mentioned in the comments).
UPDATE 2019
Looking back at this answer, a better suited choice would be the below-mentioned mersenne twister. Or you could find any implementation of xorshift.
The Mersenne Twister may be a good pick for this. As you can see from the pseudocode on wikipedia, you can seed the RNG with whatever you prefer to produce identical values for any instance with that seed. In your case, the maze ID or the hash of the maze ID.
If you are using Python, you can use the random module by typing at the beginning,
import random. Then, to use it, you type-
var = random.randint(1000, 9999)
This gives the var a 4 digit number that can be used for its id
If you are using another language, there is likely a similar module

Resources