Dice Rolls Issue



  • There is definitely an issue with the algorithm used to randomize the Dice rolls. Very often a game will result in spitting the same numbers again and again. I recently just played a game that rolled a 9 almost every time. And when it didn't it rolled a 7. It was crazy.It seems to be a different set of numbers per game. As if for each game, the algorithm decides on 2-3 numbers to roll very, very frequently.

    Might be something worth looking into.



  • I can agree with the OP. I, BobPlop, was playing a game with Jgill84 against two Rookie AIs. Both of us are in browser. I rolled a 7 five times in a row, and AI rolled a 7 too. This is more than just luck



  • So I am not discounting my previous post, nor OPs, but people should look at the statistics at the end before posting. In fact, screenshots of the dice roll graphs should be mandatory. The reason I bring this up is because I just got through a game of me, gf, and 2 rookie AIs and we swore repeatedly that the 7s bug was in effect. Like a 7 every other turn. We soldiered through it and the game eventually ended. I went to the stats at the end

    After research I found that to my surprise, the rolls were actually pretty even based on their probability. The dreaded 7 was rolled 18x on 85 total dice throws, 21% of the time. In the probability chart, it should have been rolled 14x (17%). That is not far off at all, def. not statistically significant.

    Probablity Chart:
    http://wizardofodds.com/games/images/craps/dicetotals.gif

    Actual game play chart:
    http://i.imgur.com/lWNdUtk.png

    Rest of Estimates vs Actual rolls and probs. Notice how they are all pretty much in line.
    http://i.imgur.com/0zM6PkO.png



  • Probabilities over a small set of numbers does not always follow normal distribution rules like that. 50-70 rolls is not enough to show a good distribution.

    If the dice prefer one player more constantly, especially AI... That might be a problem but we can not be sure of that.



  • 50-70 rolls should be plenty to see if there is a number that is being rolled more or less overall. But you do bring up a good point that it is not justt the overall distribution, but each persons roll distribution. thats not a bad idea, from the devs side. That level of detail doesnt show in the end stats though



  • 50-70 rolls over 10 different numbers... If you flip a coin 7 times, you can easily have 5 heads and 2 tails. And it is OK.



  • So for 10 numbers, what would be the ideal number of throws to determine if the dice algorithm is fair? I still think 85 is more than enough to get a good idea.

    I still think the 7 is an issue though. We stopped a game because I got a 7x multiple throws in row and other AI got a 7 too. I wish there was a way to get the stats without spending an hour to complete the game. I also wish it showed each persons dice throws. Because, overall, the 7 may have been statistically even, but I was getting it every time.



  • The way the statistics and probabilities work... You would need at least 100 rolls per number so... 1000 rolls?

    7 is the most common number so it rolling a lot is expected. Also, no matter what your previous rolls were, the next roll is not determined by them. You may have 20x7 rolling in a row. Although it is highly not probable, eventually it will happen somewhere.

    The statistics should be much more detailed indeed. In PlayCatan you could see how many resources each player gained or lost from different sources (rolling, robbing, using dev cards). It would be good to add how many times each player got resources with each number (and how much). And you could filter it by turns or something...



  • Two more games, with 2 more roll stats at the end. No glaring 7 errors with us.

    Game 1: 8 got screwed (3 actual rolls vs 9 probable)
    http://i.imgur.com/cBGTcBs.png

    Game 2: Basically a perfect bell curve (8 had most)
    http://i.imgur.com/srdNJBQ.png



  • This is usual. Catan is not a fair game rolls wise. You have to make up for it by using the robber, trading with the right people and NOT trading with the person getting lucky rolls.



  • Can you please start posting stat rolls at the end? hard evidence is always perferred



  • So after like 10-20 games i'm beginning to believe there is a dice rolls issue. It may not be on the #7, here it is on #6, but random numbers are being rolled hard. I notice dice roll abuses significantly more using this computer game than when I play the actual board game.

    Here, in a game with 63 dice throws, the number 6 was rolled an outstanding 18x, accounting for 29% of all throws. This is twice the probable amount of 14%. As a result, #7 went down from 17% to 6%, and #8 crashed from 14% to 3%. This caused the game, 1 human against 3 masters, to be done in less than 30 minutes (seafares 14VP). The AI that had the 6 crushed everyone especially when they started putting 2 and 3 settlements on it.

    For what its worth, two days ago I was playing in a game that had 97 throws. #7 was rolled 19x and #8 was 21x, thats a combined 42% of all throws. Probable amount should be 31%. Now 11% difference may not sound like a lot, but it had a huge (negative) impact on the game.

    Catan screenshot
    http://i.imgur.com/0aWKXI1.png

    Excel chart of probable/actual values
    http://i.imgur.com/HjsCORZ.png


  • administrators

    @BobPlop Thanks for the time you put in your observations but there is one detail i want to note. If observing statistics of dice roll (wih the porpuse of evaluating the distribution) you can´t do that one game at a time. You will get resultes that are all over the place and therefore you won´t be able to make any informed statemaent about the distribution.

    One roll of dice that has been made doesn´t affect the one that follows. This goes for games as well. Pleople tend to believe that after a long time a certain number hasn´t been rolled the probability rises that this number is rolled. The truth is that the porbability alwayas stays the same.



  • Right, each dice throw is independent of each other. This isn't a problem in real life where dice throws aren't simulated. In catan, dice throws are being simulated, and the simulation is off at times. Sometimes, like my experience above, it is way off and it ruins the game.


Log in to reply