Pythium revelers

[JimMorrison]

Well, I just ran a test batch, 9 nations (well 10, but Midgard has no mages with random picks to test, I did not know this!), trained 20 mages each, tried to get some variety in pick style-

Arco - Mystic, 3 50% picks: 3 got 0 (15% with 12.5% probability), 3 got 1 pick (15% with 37.5% probability), *13* got 2 (65% with 37.5% probability.), 1 got 3 (5% with 12.5% probability).
Marignon - Goetic Master, 1 10% pick: 2 got the pick, 10%.
Jomon - Kanushi, 1 10% pick: 1 got the pick, 5%.
C'tis - Sauromancer, 1 10% pick: 1 got the pick, 5%.
Utgard - Norna, 1 10% pick: 1 got the pick, 5%.
Patala - Nagarishi, 1 10% pick: 4 got the pick, 20%.
Atlantis - Angakok, 1 10% pick: 2 got the pick, 10%.
Pythium - Reveler, 1 20% pick: 6 got the pick, 30%.
Bogarus - Skopets, 1 20% pick: 4 got the pick, 20%.

This leaves us with an average fulfillment of (222%/250%) = 89% of random pick consistency, on a test body of 180 mages.

<3

[PvK]

Well within standard deviation, no?

[Lingchih]

Well again, to harp back on an idea that has been posted before, I think the random engine is flawed. I am not a programmer, but I do know how random engines work, and some are better than others.

The Dom engine, as I have witnessed through many different events, seems to hone in on it's first random number. I have gotten six of the same type of tartarians in each consecutive turn in a game, and received several of the same events time after time in games.

That said, I know that truly random engines are very hard to code. I think the one in Dom 3 works fine most of time.

[JimMorrison]

PvK said:
Well within standard deviation, no?

Indeed. The smaller samples show larger swings of course, and the larger the sample, the more homogenized the picks begin to look.


As far as the RNG goes, I'm not totally sure I agree that there is some fundamental flaw beyond it being an RNG. I mean, there are only 5 different types of Tarts, right? And you got 6 of one type in a string, which is relatively unlikely. But have you ever played Yahtzee? Ever been dealt a straight flush or 4 of a kind in poker? If you saw that sort of blatant streaks very commonly, I am sure that it would be indicative of a problem.

Really though, random is different from chaos. It would be interesting if there was a chaos generator rather than a random generator, and it looked at what you had gotten, and increased the probabilities on a geometrically curved bias skewed towards what has been lacking. Unfortunately, doing so would render ultra high prot/MR/def to be even more powerful than it already is, because rather than having a 16% chance to roll another 6 after your first one, the probability would artificially drop. Bear in mind, your odds of rolling a second 6 are NOT 2.8%, they are in fact still 16%. This is why the general rule of thumb will always be - short term streaks, long term statistical averages.

[Karlem]

Regarding random generator seed and such.

I think that it's decided when you create a game, why? I've seen this multiple times:

1.- Create a game, play a couple of turns
2.- Back up the game.
3.- Host it. Random events X & Y.
4-. Use back up and host it with no changes. Same random events X & Y.

Repeat point 4 a couple of times and that should prove this, isn't it?

[thejeff]

True, the seed probably is decided when you create a game, but that has no bearing on whether it's a good psuedo random generator or not.

Rerun a turn with the same orders, everything should come out the same.

Rerun a turn with even one slight change (that requires a different random roll) everything after that should be different.

[sector24]

Two important things to note:
1) The N2 reveler is garbage due to Epoptes, all that matters is the blood random, so it's really a 10% chance to get what you want.

2) The average is not as important as the worst case scenario, because you really only need 1 reveler (2 would be nice but is unnecessary). When the worst case is 50 or 100 recruits before you get a blood random, that sucks.

I think this is an unusually big deal for LA Pythium because the satyr you summon is extremely useful. Not just as an assassin, but as an assassin with leadership 40 that can summon maenads when there's nothing better to do. An extremely versatile commander all for a single blood slave. They are head and shoulders above an indy commander, and cheaper as well (both initial cost and upkeep). To not take advantage of this is foolish, but at the same time to try and take advantage of this is also foolish.

[MaxWilson]

JimMorrison said:
As far as the RNG goes, I'm not totally sure I agree that there is some fundamental flaw beyond it being an RNG. *snip* Really though, random is different from chaos. It would be interesting if there was a chaos generator rather than a random generator, and it looked at what you had gotten, and increased the probabilities on a geometrically curved bias skewed towards what has been lacking.

A high-quality pseudorandom number generator will produce numbers in the observation domain that are statistically uncorrelated with each other. That's about as good as you can do with stateful generators. It would not surprise me if Dominions had a weak pseudorandom number generator (JK is unlikely to have written one himself). I don't call that a bug, but it does mean I'm skeptical about certain kinds of empirical testing. Instead of one person recruiting 1000 Revelers and counting the random picks, I'm more inclined to trust the average of 10 people each recruiting 100 Revelers (in separate games) because the Revelers picked all on one machine may be correlated.

-Max

[Endoperez]

Reported problems with the Revenant randoms:
Coldshard, ano, llamabeast


Revenant random distribution reported as working fine:
Ironhawk, Lingchih, llamabeast, Wick, Endoperez, Jim Morrison

It's not quite 10, but it's close.



Also, one thing to keep in mind when speaking about RNG is that the computer doesn't categorize things. Things are just things. For a human, finding four sites with fire mages in your provinces seems like too small of a chance to be a coincidence. For the RNG, it's provinces A, D, E, F, H, L, M, N and P having, among others, sites 100, 185, 24, 280 and 65, or something. RNG problems with sites probably aren't visible to the player, unless multiple provinces have the exact same provinces or the same site appears multiple times in a province.

[MaxWilson]

Endoperez,

I agree about the categorization. That's why finding a bunch of illusionists in one game is just weird (and possibly observer bias) and not necessarily a bias in the game. I *wonder* about the quality of the pseudorandom number generator, but without knowing the implemenation it's hard to know if the pseudorandom number generator could make any difference. Suppose that one particular seed is more likely to generate 2's on a d8 roll. Does that translate to more Air sites than normal? Only if d8 rolls are involved in site generation, e.g. if the game rolls for the chance of a site existing and then randomly picks which path it will be from a path mask. But it could just as easily do it the other way around (pick the site directly, not the path first) and in that case the d8 weirdness wouldn't matter.

So. It's not too small to be coincidence, but I've just seen things occasionally that make me wonder.

-Max