Under what circumstances that you would want to have a different seed for you random number generator?

Question

I am wondering under what circumstances that you would want to have a different seed for you random number generator. I see code somewhere that people create a Random object every time when needed, and also see people sometimes would have it as an instance variable that uses a single seed when the object is constructed. For example

// same seed
class A {
    private Random random;
    public A() { random = new Random(); }
    public int nextInt() { return random.nextInt(10000); }
    public double nextDouble() { return random.nextDouble(); }
}

versus

// different seed is used every time when you call the method
class B {
    public int nextInt() { return new Random().nextInt(10000); }
    public double nextDouble() { return new Random().nextDouble(); }
}

Please ignore this design for these two wrapper classes, just trying to show the difference.

Answer 1

Two reasons to re-use a Random:

• There is an overhead to creating a Random.

• If code is fast, creating two different Random objects may end up using the same seed value, i.e. result will not be random. Even if code is not that fast, the different Random objects will be seeded with closely related seed values, reducing the randomness factor.

One reason for having multiple Random instances, though not a reason for creating a new one on every use. Quoting javadoc:

Instances of java.util.Random are threadsafe. However, the concurrent use of the same java.util.Random instance across threads may encounter contention and consequent poor performance. Consider instead using ThreadLocalRandom in multithreaded designs.

Even in a single-threaded application, using ThreadLocalRandom may be appropriate, as a simple means of obtaining a shared Random instance.

Answer 2

A pseudorandom number generator, such as java.util.Random, produces a deterministic sequence of numbers characterized by the initial seed. Such generators are crafted to produce sequences that appear to vary randomly and (usually) uniformly over the whole range of the PRNG.

You get the same sequence of numbers if you start with the same seed. That can be useful under some circumstances, but usually it is undesirable. In particular, your numbers will not have any appearance of randomness at all if you seed the PRNG with the same seed before generating each number.

But more subtle problems arise even if you seed the PRNG with different seeds before generating each number, including if you instantiate a new Random instead of reusing an existing one. In this case, you are not guaranteed to have the same range or distribution of results, and your results may exhibit correlations that they would not exhibit if you seeded only once.

So, when you say

I am wondering under what circumstances that you would want to have a different seed for you random number generator.

, about the only reason I can think of for wanting to change seed within one run of your program is that you have published enough draws from the PRNG that you are concerned about someone determining its implementation and internal state, and therefore being able to predict subsequent numbers. This would be a concern in practice only under rather special circumstances.

On the other hand, you generally do want to choose a different seed each time your program runs, else it will generate the same sequence of random numbers each time.

Generally, you should seed your PRNG once, and then use it as much as needed, without reseeding. It's not necessarily wrong to use multiple, independent PRNGs, but the reasons for doing so are mostly organizational (e.g. not having to worry about synchronization, not having to provide for injecting a Random instance) rather than functional.