Luck/Turmoil versus Nothing.

[Squirrelloid]

Originally Posted by chrispedersen

I still don't get the criticism.

I'm suggesting first determine how many provinces get event checks. Then run each event check through a provinces event mask.
But I'll do some tests...

Ok, two things:

First, as thejeff pointed out, if you determine #events before you look province-by-province, then province scales have no effect on how likely an event is. Yet they clearly claim to do so.

Second, my point was if you generate #events first, you ignore #provinces. Yet there's certainly the perception that larger empires have more events (we need real data to actually confirm, but it does meet my anecdotal experience). At which point, generating #events first doesn't account for increasing average # of events based on number of provinces owned that doesn't result in obvious asymptotes or the like at some point. (And remember, JK's known algorithm work has already rejected x^2+y^2=z^2 as too complicated).

[chrispedersen]

ok; With 50 turns, 0 luck 0 luck -3 +3 4 territories
22, 27, 54, and 46 luck events

With 8 territories same scales
24 24 64 55 luck events.


A few observational data.

1. With over 100 turns run - there were never any turns with
ZERO luck events (after turn 1).
2. I watched various emigration events where pop was lost.
There were never any pop gains the same turn.
3. Four events occured to one nation once in both sample sizes.


Number distribution
With 4 territories
0,1,2,3,4
33,11,4,1
26,19.4
15.17.11.5.1
19,18,7,5



With 8 territories,
0 1 2 3 4
28,19,3,0
29,18,3,0
10,16,29,3,1
11,24,13,2

[Finalgenesis]

I will admit to not being able to fully understand this. So all tests below were with 4 nations each? When you say 22,27,54,46 luck events, you mean for the 4 nations seperately over 50 turns? For the ones below, 0,1,2,3,4 are events and the 4 rows are for 4 nations? the 26, 19.4 is that over 50 turns 26 had no event, 19 had 1 event and 4 had 2 events?

Originally Posted by chrispedersen

ok; With 50 turns, 0 luck 0 luck -3 +3 4 territories
22, 27, 54, and 46 luck events

With 8 territories same scales
24 24 64 55 luck events.


A few observational data.

1. With over 100 turns run - there were never any turns with
ZERO luck events (after turn 1).
2. I watched various emigration events where pop was lost.
There were never any pop gains the same turn.
3. Four events occured to one nation once in both sample sizes.


Number distribution
With 4 territories
0,1,2,3,4
33,11,4,1
26,19.4
15.17.11.5.1
19,18,7,5



With 8 territories,
0 1 2 3 4
28,19,3,0
29,18,3,0
10,16,29,3,1
11,24,13,2

[chrispedersen]

Originally Posted by Finalgenesis

I will admit to not being able to fully understand this. So all tests below were with 4 nations each? When you say 22,27,54,46 luck events, you mean for the 4 nations seperately over 50 turns? For the ones below, 0,1,2,3,4 are events and the 4 rows are for 4 nations? the 26, 19.4 is that over 50 turns 26 had no event, 19 had 1 event and 4 had 2 events?

1. No. 50 turns were done for 4 nations with 4 provinces.
50 turns were done for 4 nations with 8 provinces.
2. correct.
3. No. In the second data I am presenting the number distribution. Nation 2 had 26 turns with 0 events, 19 with 1, 4 with 2.

[thejeff]

Yes, the order they were listed in the messages.

Do we know the base chance? We certainly don't know what it means.

At least by the end of that test most provinces were T3L3, which should have boosted the chances by 36% (Do we know if that is +36 to the base percent chance, or an increase of 36% of the base chance? I'd always assumed the former...)

I can try to run more turns. See if the data changes.

[Squirrelloid]

I'm assuming its a % increase in base chance, because adding 36% is very clearly too large.

My tests suggest a 10% chance of an event for each of 4 independently calculated events. The distribution of events i saw matched such a binomial distribution. Of course, I only tested with one province, so it does leave open how it works with more provinces.

[Finalgenesis]

From a design point of view, I would think that a diminishing effect on event frequency for additional province to be a reasonable design decision, along with some sort of cap. It looks like the event gen mechanic doesn't check for event province by province, but perhaps roll for whether an event happens, then roll where it occurs among your provinces, then select events up or down the event list base on local luck. The formula itself will probably account for province number as a factor in a diminishing formula, e.g. 4 rolls of "Occurence = 10% + (10% * (x/x+1))" where X is the # of province your have (disclaimer: this is not meant to guesstimate the actual formula, but show how a cap of 19.9999% event occurence in 4 checks might be implemented, along with adding in province # as a dimishing factor, the cap can easily be changed to other percentage).

It is also a reasonable mechanic for event gen formula may also make a 1st event more likely to occur then a 2nd event, though Squirreloid's testing seem to refute this.

Linear relationshop between event frequency and province # wouldn't make sense (Also refuted by thejeff's data, though more sample population would help as 10 turns of data is a bit low, it does however, show that something unintuitive is going on with event generation though)? We know that province # is likely factor from Squirreloid's test I think, as we do see a lot more then 36% event when we had more provinces, also Squirreloid have tested this for quite a few turns, so turn progression may not influence event chance (at least for a solo province). What other factor can account for thejeff's data where the event chance seem to pick up with no change in province #?

@thejeff
For your test, was your dominion spread covering all your provinces? Or was it just spreading out from capital? Or maybe turn # is a factor when there are mutliple provinces... Thanks for your testing data too thejeff, very handy.

[Finalgenesis]

Rereading my post above, I realized that unless you're my clone, it won't be easily understandable. I'll try to lay out my thought process in an easier to understand manner:

Event generation mechanic:

a) 4 single checks for event, roll once only (not once per prov)

-At current, this looks the most reasonable, as thejeff's testing likely show that the check is not once per prov (otherwise he'd max out 4 events a turn, or close). Likely the formula used includes province # as a factor somehow in a dimishing manner, as we can say with some confidence that we DO see more events when we had more province.

b) assign all events to random province

-Of those 4 checks, any success rolled (event generated) must be assigned to a province. Technically this should be random, though many have noticed that this tend to cluster.

c) once event assigned to province, apply luck

-So an event has been assigned to a province, luck have to apply somehow. I'm guessing the event is generated after the province has been selected. The province applies its local luck to the event roll through whatever formula.

thejeff's phenomenon:

This refers to why thejeff's testing seem to show that events occurence picks up as the turn progress. To be fair we need more data to say anything with any credibility, let's leave this aside for now. Something must be different during his t5-9 compared to t10-15 to account for the increased occurence, off the top of my head:

a) Turn #

Well obviously turn # is most visible, would turn progression be a factor in event generation? Squirreloid's testing seem to say otherwise, since his average event frq on a single province was 36% throughout without seeing thejeff phenomenon. However, it is remotely possible that turn # as a factor only kicks in when there is more then 1 province. I see this as unlikely but possible.

b) dominion

The other thing that could be different is the dominion, but I need to know whether thejeff's data had dominion spreading from Cap on T1 only before I can make any conjecture. Test with philosopher and temples may be revealing if so, maybe in thejeff's test scenario wall in dominion so it only occurs at capital, then letting it max or keeping it low to see, then maybe test if dominion being widespread changes anything.

c-z) ???

[Squirrelloid]

Except there are substantial problems with it checking for each event on a non-province-by-province basis. Namely, no obvious way you accrue more events as #provinces increases, and no way to factor in local scales. So i'm not sure your (a) is anywhere near certain - indeed, it would seem the evidence suggests against it.

Now, actual P(event) could actually be P(event|#provinces), as a decreasing function with respect to provinces. This seems a rather complicated model given what we know of JK's coding style, however. (I mean, the game uses a distance metric near manhattan metric because calculating real distance was too much processor power per calculation - clearly simple was better as far as JK was concerned). But we don't *know* unless someone wants to do some code-diving.

Until we determine a plausible mechanism for generating events, can we restrict ourselves to totally neutral dominion? Lets not complicate things before we have a good basic model to work with.

[Finalgenesis]

I thought it might be pretty simple to code in #province affecting event even if you're doing a non-province-by-province basis. something like the example I used 2 post back (I didn't put the example in the "clarified" post you read so you probably missed it):
event=10%+(20%*(x-1/x)) where x is province you have
to demonstrate a dimishing effect as thejeff's data might suggest. so 1 province = 4x 10% check, 2 provinces = 4x 20% check, 3 province = 4x 27% check ...etc with this random non-realistic example (it's a pretty inelegant sample formula), except there is obviously more factors in the formula then is present in this example, in order to account for anomalies for Jeff's results.

I agree though that nothing short of code diving is going to give us anything concrete to work with, so it's an excercise in futility at the end.