Is Collections.shuffle suitable for a poker algorithm?

Question

Hi,

there is a poker-system in java, that uses Collections.shuffle() on all available cards before the cards are dealt.

So a collection of 52 cards 2-9, J, Q, K, A in 4 types.

After that we Collections.shuffle().

The problem is, that it seems (until now we didn't have big statistic, it's possible that we only see a lot of statistic inferences), that the algorithm is VERY unclearly.

So, is Collections.shuffle() okay for a poker algorithm?

---

Answers to comments: With "unclearly" I mean it's very very mysteriq at some time. Much User complain about "it's not the same as live / other pokerrooms". I played a much with this system and must say, I agree, I see 3 Royal Flashs in under 2000 played hands in this system and live/in other pokerrooms with over 100.000 played hands I see 2 until today.

Answer 1

Ok, I generally hate people saying this to me but yes and no. It's about as ok as pickrandomcardbetween(1, 52) and use a rand() function when it comes to randomness.

The no part is that for anything that deals with chance or random values you need proper hardware, a normal computer cannot even remotely generate a truly random result of any kind.

Edit: If your pokersystem is for playing for fun it's one thing, but when money is involved people will hang you for creating random results that way.

Answer 2

If this is a serious poker application, where money can change hands, the short answer is NO. For something like this, you should really use a hardware source of true randomness.

The slightly longer answer is: if you can't get hardware for doing true randomness, Collections.shuffle(List, Random) might be good enough if you supply a SecureRandom. The tricky part with this solution is finding a good seed value.

UPDATE: Based on your clarification, I'd suggest you look into how you're seeding the PRNG (assuming you're already using a cryptographically secure implementation; if not, do that first). You should not be using a limited set of seeds. Other things to consider:

• you should probably instantiate a single PRNG for each game
• you should only be shuffling the deck between hands; from your question, it's not 100% clear that you aren't also shuffling the deck between the flop, turn, river, etc.

Answer 3

The Collection.shuffle uses the O(n) implementation of the Fisher-Yates shuffling algorithm.

And the random indexes are chosen with the normal PRNG of Java, so it will be approximately uniform: every shuffle of the deck will be as much probable as every other one.

This is quite ok for what you want to do, but when you want real randomization you should introduce some real random factors (like System.currentTimeMillis() used to seed the random number generator) or something more realiable like a specialized hardware.

Answer 4

If this is serious poker software involving money, then the answer would be no. (For this, you would want some source of true randomness.) However, for simple circumstances, it is just about as good a solution as any other algorithm.

If you want more information about the shuffle algorithm itself, see http://stackoverflow.com/questions/2249520/javas-collections-shuffle-is-doing-what.

Answer 5

I suggest reading this article:

How We Learned to Cheat at Online Poker

The authors looked at one software package and found several flaws. One serious problem was the seed. If you start with a 32-bit seed (and don't generate a new independent seed during the shuffle), you can only generate 2^32 different random sequences. There are 2^226 possible shuffles of a 52 card deck, which means only a small fraction of possible deck orderings will be produced.

A player knows 5 of the card positions (7 in Omaha) at the flop. If the player knows the shuffle algorithm he can guess what the candidate seeds were based on the cards he sees. This gives him a big advantage in deducing the probabilities of what the hidden cards are.

Answer 6

The problem is that the numbers generated by random are statistically random. It means that shuffle do not behave like a deck of cards because it is more random than a real life deck shuffle. To have something more realistic you need to simulate the way you shuffle the cards in real life, like how many times you cut and so on. I saw a site with a chart comparing the results of a real dice and computer generated results that showed how different the results were. The computer results were more equally distributed, but I can't seem to find the link on google.