G-Minor XI

[Denniz]

That strikes me as a dangerous set of assumptions given what we know. We'd need to know the relative value of time before suggesting raze and replace with confidence.

We had to replicate the final score calculation. This is the formula the parser uses:

// final score calculation
$progress = floor(100 *$data['VictoryTurn'] / $data['MaxTurn']);
if ($progress == 0) {$progress = 1;}
$data['FinalScore'] = floor(($data['BaseScore'] * 100) / $progress);

I hope that helps.

 

[Martin Alvito]

I appreciate you providing that, but as far as I can tell crafty has indeed accurately reverse engineered that. If I'm being obtuse and missing something, please let me know.

The point I was making is that the algorithm generating the term 'BaseScore' remains unknown, and that it's hard to maximize BaseScore if I don't know where it comes from. Compare to Civ II where the scoring system was known, transparent and linear. If the underlying function is fully linear here, what's up with the extra parameters in the .xml? I don't really think that any term in the scoring function is an exponential, but that's an example of the ways in which the unknown score function could be deviating from linearity here.

 

[vexing]

starting a new game, everyone starts with 4 points (agriculture)
planting city adds 19 points: 8 for city, 4 for 1 pop, 7 for tiles.
getting a wonder adds 25 points.
everything is simply * multiplier... i have no clue what the "factor" things in the defines go into, if anything.

if you highlight your score in game, it shows a breakdown of your points from each category.

i'm not sure i'll even enter this one, i guess it'll depend on how much free time i have next week.

 

[Martin Alvito]

OK, so the last six terms are linear.

All we need to do, then, is figure out what the FACTOR items are. If I had to guess, they're caps or triggers that modify the rate of score gain above those values. Probably the best way to figure out what those do would be to use FireTuner and see if a massive civ with known parameters still has a score that behaves linearly.

I'll try to fiddle with that later today if I get some time.

 

[Sporaticus]

A malfunction in the systems of the stealth bomber has created some damage
You have to be creative

Aha! The defense contractor, who just happens to be the 2nd cousin of the chairman of my civ's defense appropriations committee, has been using those substandard rivets again! Time to reinstate the Monarchy.....

This particular G-Minor is turning out to be quite fun! I'm starting to develop a feel for how to "game" the game. Since I'm an "instinct" player rather than a "formula" player, it is hard to put into words. There seems to be a balance between working the City States, working the RA's, and ambitious expansion that creates a substantial advantage over the AI. Maybe after another 30 games, I'll be able to develop the type of drooling, ruthless, bloodthirsty, carnivorous nightmare of an uber-strategy that I need to compete in a G-Major. I just hope the patch-happy devs can contain themselves for a while.

 

[Martin Alvito]

It isn't linear, but it looks close.

I set up a game with a 50x50 map. There is a 13x13 grid of cities on the map; 168 of them belong to Rome, and one belongs to India. 2497 tiles belong to Rome. Three belong to India. Rome has all basic techs and no future techs.

Now, I should observe the following values here:
Cities: 168 X 8 = 1344
Population: 168 X 40 X 4 = 26880
Land: 2497 X 1 = 2497
Wonders: 0
Technologies: 73 X 4 = 292

Instead I get:

Spoiler :

Huh? India's off too by a point, by the way. India should have 4 for Agriculture, 8 for the city, 4 for the one point of pop, and 3 for tiles yielding a score of 19. Instead they have a score of 36. However, I loaded that screenshot on Deity, and the four bonus techs account for 16 points. Sure enough, if I play the India side, I get an extra point for the city for a score of 20 instead of the expected value of 19.

Now, this isn't the end of the world because the multiplier appears to be between 1.2 and 1.25 in each case for Rome, so the function is approximating linearity closely enough for this purpose. The game reads the map size as Small, which could explain the odd multipliers on score. It may not fully pick up in the India score due to rounding; only the city value is large enough to get hit by the multiplier for a full point, so if fractions are dropped rather than rounded or kept and added, that would make some sense.

 

[pholtz]

First attempt. Found myself isolated on a peninsula, and no iron. Ah well, attempt number 2 coming up.

 

[Peets]

First attempt. Found myself isolated on a peninsula, and no iron. Ah well, attempt number 2 coming up.

Are you going for fastest date?
Otherwise I see no reason to start again, it's the easiest level...

 

[Mallow]

what a crap... turn 524 I broke the science... Future tech -121564 of 14107 and it tells me that I have 7 turns to go

 

[sanabas]

what a crap... turn 524 I broke the science... Future tech -121564 of 14107 and it tells me that I have 7 turns to go

Yeah, I did that on the time game I played. I think it let me get to ~200k unused science before ticking over to negative values. Number might be off, since 2^17 is 131k, and 2^18 is 262, and I'd assume it's one of them. Need to either make sure your science is only just over the threshhold for 1 turn research so the excess doesn't grow big enough, or save 1 SGL for every time it does.

Haven't tried this gauntlet yet, but one of the things that the city governor was doing in my time game was repeatedly using unemployed citizens and leaving empty specialist slots.

 

[vexing]

Number might be off, since 2^17 is 131k, and 2^18 is 262, and I'd assume it's one of them.

it's 2^31 / 1000, standard integer overflow.

i had previously reported this in bug reports, if you want to confirm it: http://forums.civfanatics.com/showthread.php?t=421544

 

[sanabas]

it's 2^31 / 1000, standard integer overflow.

i had previously reported this in bug reports, if you want to confirm it: http://forums.civfanatics.com/showthread.php?t=421544

Hehe, I'm a maths geek, not a programmer.

My memory is out by an order of magnitude, too, since that's 2.1 million, not 210k.

Actually, looking at your bug thread, it appears my memory's ok, and it's actually 2^31/10000.

 

[vexing]

Hehe, I'm a maths geek, not a programmer.

My memory is out by an order of magnitude, too, since that's 2.1 million, not 210k.

i'm both =) and woops, i didn't even bother checking the actual numbers, i just assumed it'd be / 1000.

 

[Martin Alvito]

That overflow bug is going to generate some seriously amusing equilibrium behavior.

 

[killerloop]

Martin Alvito wrote:

That overflow bug is going to generate some seriously amusing equilibrium behavior.

Right, and as I already experienced this bug several times I will not enter this gauntlet.


The other interesting question is, when (going for score) is the right time to stop and win with the HOF multiplier outweighing the in game score increase possibility.

At the end of the game:
-score only increases from Future techs and growing Population
-all tiles (except the ones for the maritime states - they still help growth) and the 1 city should be yours.
-cities should be maxed out on the map.
-all wonders & techs you should have
-at the turn you finish domination you should also grab all maritime CS's to get those tiles.

With Crafty's HOF calc formula and with the info you have from your map (satelites) you can now estimate when stopping pays out. My initial calcs show that on a pangea standard map, it's better to play till the end to increase score....,

....which will not happen as we have a BUG that is known for ages.

 

[sanabas]

Each time you trip the science overflow to negative, you lose maybe 60-80 points from 6-8 future techs, or 10 points if you have a great scientist handy. The final result may end up that close, but I doubt I'd abandon a game because I was 200 points down on what I expected thanks to that.

 

[Sporaticus]

Well, the game-runs-too-doggone-slow factor kicked in at turn 600, so 15K is the best I'm going to do on this one. I purposely held back on the high-end research structures, so I did not encounter the nasty tech overflow bug.

 

[Martin Alvito]

I don't think that holding back on science buildings is the answer, but you'll need to anticipate and react accordingly.

I'm also dubious on entering this one due to turn times. I just had a Deity game grind to a halt (20-30 seconds) in between turns at turn 140 yesterday, and I don't even want to talk about what it was like by 160. I suppose that there are ways to keep that from happening given that it's Settler and the problem seems primarily to be AI decision-making for its unit spam, but if things get ugly I'm very unlikely to finish.

I'm not entirely convinced that playing all the way to turn 750 is the answer. Losing 1/100 of your score eventually gets pretty painful as your score gets large.

 

[vexing]

I'm not entirely convinced that playing all the way to turn 750 is the answer. Losing 1/100 of your score eventually gets pretty painful as your score gets large.

the answer is to keep track on a turn by turn basis and end the game when/if you start losing final score points.

 

[Martin Alvito]

Right. A limit exists; I'm just theorizing that it arrives before 750 turns.