Trying to produce a unique sequence of random numbers per iteration - c++

Question

As the title states, I'm trying to create a unique sequence of random numbers every time I run this little program.
However, sometimes I get results like:
102
201
102
The code
#include <cstdlib>
#include <ctime>
#include <iostream>
using namespace std;
int main() {
for (int i = 0; i < 3; i++) {
srand (time(NULL)+i);
cout << rand() % 3;
cout << rand() % 3;
cout << rand() % 3 << '\n' << endl;
}
}
Clearly srand doesn't have quite the magical functionality I wanted it to. I'm hoping that there's a logical hack around this though?
Edit1: To clarify, this is just a simple test program for what will be implemented on a larger scale. So instead of 3 iterations of rand%3, I might run 1000, or more of rand%50.
If I see 102 at some point in its operation, I'd want it so that I never see 102 again.

Answer 1

First of all, if you were going to use srand/rand, you'd want to seed it once (and only once) at the beginning of each execution of the program:
int main() {
srand(time(NULL));
for (int i = 0; i < 3; i++) {
cout << rand() % 3;
cout << rand() % 3;
cout << rand() % 3 << '\n' << endl;
}
Second, time typically only produces a result with a resolution of one second, so even with this correction, if you run the program twice in the same second, you can expect it to produce identical results in the two runs.
Third, you don't really want to use srand/rand anyway. The random number generator in <random> are generally considerably better (and, perhaps more importantly, are enough better defined that they represent a much better-known quantity).
#include <random>
#include <iostream>
int main() {
std::mt19937_64 gen { std::random_device()() };
std::uniform_int_distribution<int> d(0, 2);
for (int i = 0; i < 3; i++) {
for (int j=0; j<3; j++)
std::cout << d(gen);
std::cout << "\n";
}
}
Based on the edit, however, this still isn't adequate. What you really want is a random sample without duplication. To get that, you need to do more than just generate numbers. Randomly generated numbers not only can repeat, but inevitably will repeat if you generate enough of them (but the likelihood of repetition becomes quite high even when it's not yet inevitable).
As long as the number of results you're producing is small compared to the number of possible results, you can pretty easily just store results in a set as you produce them, and only treat a result as actual output if it wasn't previously present in the set:
#include <random>
#include <iostream>
#include <set>
#include <iomanip>
int main() {
std::mt19937_64 gen { std::random_device()() };
std::uniform_int_distribution<int> d(0, 999);
std::set<int> results;
for (int i = 0; i < 50;) {
int result = d(gen);
if (results.insert(result).second) {
std::cout << std::setw(5) << result;
++i;
if (i % 10 == 0)
std::cout << "\n";
}
}
}
This becomes quite inefficient if the number of results approaches the number of possible results. For example, let's assume your producing numbers from 1 to 1000 (so 1000 possible results). Consider what happens if you decide to produce 1000 results (i.e., all possible results). In this case, when you're producing the last result, there's really only one possibility left--but rather than just producing that one possibility, you produce one random number after another after another, until you stumble across the one possibility that remains.
For such a case, there are better ways to do the job. For example, you can start with a container holding all the possible numbers. To generate an output, you generate a random index into that container. You output that number, and remove that number from the container, then repeat (but this time, the container is one smaller, so you reduce the range of your random index by one). This way, each random number you produce gives one output.
It is possible to do the same by just shuffling an array of numbers. This has two shortcomings though. First, you need to shuffle them correctly--a Fischer-Yates shuffle works nicely, but otherwise it's easy to produce bias. Second, unless you actually do use all (or very close to all) the numbers in the array, this is inefficient.
For an extreme case, consider wanting a few (10, for example) 64-bit numbers. In this, you start by filling an array with numbers from 264-1. You then do 264-2 swaps. So, you're doing roughly 265 operations just to produce 10 numbers. In this extreme of a case, the problem should be quite obvious. Although it's less obvious if you produce (say) 1000 numbers of 32 bits apiece, you still have the same basic problem, just to a somewhat lesser degree. So, while this is a valid way to do things for a few specific cases, its applicability is fairly narrow.

Answer 2

Generate an array containing the 27 three digit numbers whose digits are less than 3. Shuffle it. Iterate through the shuffled array as needed, values will be unique until you've exhausted them all.
As other people have pointed out, don't keep reseeding your random number generator. Also, rand is a terrible generator, you should use one of the better choices available in C++'s standard libraries.

Answer 3

You are effectively generating a three digit base 3 number. Use your RNG of choice to generate a base 10 number in the range 0 .. 26 and convert it to base 3. That gives 000 .. 222.
If you absolutely must avoid repeats, then shuffle an array as pjs suggests. That will result in later numbers being 'less random' than the earlier numbers because they are taken from a smaller pool.