Feature request: true random dice

[djensen]

I’ve played multiple dozens of 1942 SE maybe hundreds? I can count the number of times I lost an attack on the UK fleet on turn one on one hand. I’ve played maybe 6-12 games of Online and I’ve lost 3 times. I realize the dice are “cryptographic” random seeds but all that means is you cannot predict the random seed.

I would pay extra for true random dice. Random.org has a service for true random numbers: “The randomness comes from atmospheric noise, which for many purposes is better than the pseudo-random number algorithms typically used in computer programs.”

https://www.random.org/dice/

[redrum]

@djensen It would be interesting to know how A&A online does the randomness for the dice. TripleA uses the Mersenne Twister pseudorandom number generator which is the most widely used general-purpose PRNG: https://en.wikipedia.org/wiki/Mersenne_Twister

[JuliusBorisovBeamdog]

Hey! This has been brought up a number of times, especially at the game launch. A PRNG generates the value. We’ve examined its output versus real random data and have found it to be indistinguishable. This also applies to the Computer opponent.

[aardvarkpepper]

This post is deleted!

[Imperious Leader]

I called this nonsense before. The dice are “tit for tat” meaning if you miss he misses…if you hit he hits usually… on his way out. I lost a game ( my only real loss by three units left trying to take Moscow) in this battle i should have lost but the payoff was the game…so i went in and had great results. Then at the end he had one unit to my three ( including a fighter, Bomber, Tank to an infantry) he hit every time and i missed every time and held Moscow. I lost after 3 more rounds ( Berlin Taken)

[djensen]

On the other hand, the dice seem relatively even keel when you play a bunch of games.

Even with physical dice, I’ve had some crappy results.

[djensen]

I’m still super suspicious of the dice and this is coming from somebody with unlucky physical dice rolls.

I’m playing one game in particular where all my rolls are bad.

How does the PRNG work on the server? Do I get stuck with the same randseed for each game? Is there a new randseed before each roll? I can see where the distribution of rolls across all games might appear “normal” but are there any abnormalities that would be more rare in nature than using whatever PRNG techniques are being used?

We don’t want to be able to run an algorithm and predict the next roll. But experiencing 10 6s in a row should still only be a 1 in 60 million. I’d be curious to see if that’s happened even once and if there have been 60 million rolls to account for that.

I’m not really interested in the statistics around individual rolls en masse, I"m interested in the strings of rolls together and how much they match with probabilistic reality. This might be at the heart of why A&A players in general hate PRNGs more than they hate physical dice (which we do).

[Imperious Leader]

Do you notice that the defender nearly always hits on the way out, especially if they had bad rolls before?

I got way too many rolls where nothing hits and zero ones where most hits.

[djensen]

I’m not finding consistency or a pattern just too many very bad rolls. I played an entire weekend of Anniversary Edition with bad rolls but nothing like some of the bad rolls I’ve had on 1942 Online.

I think there are two levels of bad rolls:

“Well that truly sucked”

and

“WTF? Are you kidding me!”

In my experience, in face-to-face games, I get a lot of the first one and rarely the second one; maybe once a game or once every other game. In 42 Online. I’ll get that multiple times per game and in other games between 0-1.

[djensen]

I will admit, @redrum, that when I play TripleA with the built in dice roller; I’ve never felt like the results were “weird” or out of place. Some of the dice servers on the other hand, hmm.

Maybe it’s the Mersenne Twister that’s the difference. @JuliusBorisovBeamdog would it be possible to find out if this method for PRNG is the one being used or not?

[JuliusBorisovBeamdog]

I have asked the dev team for an additional confirmation/details.

XorShift128+ is the PRNG we use (currently a standard in a variety of environments, including v8). The value is time-seeded every roll, not per-game. In our own internal testing, we found it sufficiently random.

http://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf

[redrum]

@JuliusBorisovBeamdog There is some research that has shown weaknesses of the Xorshift128+ generator:
https://arxiv.org/abs/1908.10020
https://arxiv.org/abs/1907.03251
https://lemire.me/blog/2017/09/08/the-xorshift128-random-number-generator-fails-bigcrush/

But yes it has been adopted by some pretty highly utilized tools like Javascript 8 though the PRNG they used before was well really bad: https://hackaday.com/2015/12/28/v8-javascript-fixes-horrible-random-number-generator/

I’m not sure it really matters and to some extent you’ll always have people complain about getting screwed by dice no matter what you do.

[aardvarkpepper]

This post is deleted!

[djensen]

Um, not sure where @aardvarkpepper is going but I’m with @redrum.

The dice rolls in TripleA using the internal dice have always seemed “natural” a bad roll and a good roll here and there but nothing that doesn’t seem unnatural.

The choice of dice algorithm in AA42O is just bad. I frequently get zero hits in situations where it would be 5-10% of that happening. At this point I’m playing enough games and rolling enough dice that these statistically anomalies should be, just that anomalies.

If the team stands by their PRNG algorithm, then show me my lifetime deviation on dice rolls. I.e., what I rolled vs. what the probability was.

An maybe, I have played enough games that these bad rolls are actually meaningful. So just let me know; it’s also a fun feature to encourage the gamblers fallacy. :rolling_on_the_floor_laughing: “Red just hit on the roullette table 10 times in a row, there’s no chance it will hit agai … damn.”

It would also be cool to see the global expected vs. actual dice results as well.

[aardvarkpepper]

@djensen said in Feature request: true random dice:

Um, not sure where @aardvarkpepper is going

Yes all right, I’ve deleted my posts in this thread.

[djensen]

@aardvarkpepper Aw, you did that to go that far. :cry: I was just a little confused is all.

[djensen]

@redrum said in Feature request: true random dice:

I’m not sure it really matters and to some extent you’ll always have people complain about getting screwed by dice no matter what you do.

Yeah but still, I’ve played several fast solo games, to test out ideas, and a bunch online and my impression is overall positive. I think the best way to overcome is to show users where they stand in terms of rolls. How many and how far outside of the average are they.

[JuliusBorisovBeamdog]

I have asked around. We don’t have global statistics on the rolls, unfortunately.

[djensen]

I’ve had it with these dice. 2 rolls 4 tanks each roll, all misses. This is worse than my worse rolls IRL. I’m seriously about to rage quit all my games.

[djensen]

And my opponent roll over 50% on infantry only twice in a row.

Effing seriously HOW MANY 99.5% BATTLES DO I NEED TO LOSE? If I go into 100 battles at 99.5%, over the long term, I SHOULD LOSE ONE FOR EVERY 200.

Okay let’s get to a more significant number, 1000. I should have lost 5 battles at 99.5%. I have not had 1000 battles, I’ve lost at least a dozen of them. There’s something wrong. Even at 2000 or 3000 I’m way ahead of the curve on losses.

It’s not fun if I have to go into every battle at over 99.9%. And even then I’m probably going to lose 10 out of 1000.