Home Trending Latest threads New posts Developer diaries

leaders generator project

Jump to latest Follow Reply
  • We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
Show only dev responses
not ready to reply to TBox regarding all things.

Manus influence on leaders - i suppose more than majority refuce to improve the strength of rich and powerful nation more by leaders. Mostly such great countries have already good leaderfile.

Quality slider was not offered for logical influence on leaders by near same reason. But nice ideas to use sliders for tweaking leaders params are always welcome.
 
regarding destribution of leaders.

We`ve talked a lots with ForzaA, thanks him for his ideas about formula to implement something like gaussian destribution (which is parabolic with max value at average), instead of current destribution when every digit has equal chance - for example same chance to get shock 2 and shock 6.

So we came to formulizing like this:
current received value + rand (0..1)*(0..1) * multiplier (positive and negative), which leads to parabolic destribution and results are will be more close to the average and far from minimum and maximum.

Of course, there are some tweak required for using when average is not directly between min and max, but i hope to implement that this week.
 
How about a stickied guide with links and instructions on how to create leaders? I have put a lot of effort into understanding, sending private PMs and browsing forum posts and still no jackpot.
 
How do I point to my own save?

EDIT:

Ok I'm doing progress now.

What was really missing was the link to the stats creation page.
 
Last edited:
Tonioz said:
Of course, there are some tweak required for using when average is not directly between min and max, but i hope to implement that this week.
Click to expand...

I though that average was always between min and max ;)

I assume you want to change the mean from the theoritical one. When you choose a number in your distribution you could add a random number to it. So, it's like having a Gaussian distribution but with an effective different mean and standard deviation.
 
He's talking about the min and max for leaders.

If you're generating extremely crappy leaders, and ask for an average of 2, then run the Gaussian distribution he's showing, you'll get values between -2 and 6, and the average will be 2, as requested. But then you have to truncate all the -2 and -1 to 0, and that kicks the average to above 2. And gives you a lot of zeroes.

Correcting the math so that, after truncation, the final average is what you requested originally is what Tonioz is talking about.
 
TBox said:
He's talking about the min and max for leaders.

If you're generating extremely crappy leaders, and ask for an average of 2, then run the Gaussian distribution he's showing, you'll get values between -2 and 6, and the average will be 2, as requested. But then you have to truncate all the -2 and -1 to 0, and that kicks the average to above 2. And gives you a lot of zeroes.

Correcting the math so that, after truncation, the final average is what you requested originally is what Tonioz is talking about.
Click to expand...

oh! Thanks! :eek:o :)
 
balinus said:
I though that average was always between min and max ;)

I assume you want to change the mean from the theoritical one. When you choose a number in your distribution you could add a random number to it. So, it's like having a Gaussian distribution but with an effective different mean and standard deviation.
Click to expand...


As far as I understand it.. Tonio wants people to be able to set:

1) the minimum value
2) the maximum value
3) the average (or mean)

for example, min 1, max 6, average 4..

that can't be done with a "simple" Gaussian.. but I think/hope his next incarnation of the leader generator will have a semi-Gaussian between the min and max values with a mean value as specified by the one having the leaders generated.
 
TBox said:
He's talking about the min and max for leaders.

If you're generating extremely crappy leaders, and ask for an average of 2, then run the Gaussian distribution he's showing, you'll get values between -2 and 6, and the average will be 2, as requested. But then you have to truncate all the -2 and -1 to 0, and that kicks the average to above 2. And gives you a lot of zeroes.
Click to expand...

As far as i saw that was such problem for all existed generators (at least i suppose). Watching converting and colonizing at low chances at EU2 i tend to think that Johan did same mistake for EU2 randomizing engine.

I did special procedure to get exactly ordered average, which was tested at 10 000 cycles and always matches with very minimal gap.
 
10 000 cycles tests.
Inputs: min - 0, max - 3, average - 1.5
Previous results:
0 - 25,03%
1 - 24,04%
2 - 25,07%
3 - 25,86%
So destribution is linear - each integer has equal ~25% chances to be got.

Results after when parabolic distribution is done from middle to min and from middle to max:
0 - 15,07%
1 - 34,46%
2 - 35,92%
3 - 14,55%

Defined average and results average are still very very close, so results are very accurate with method used.

Then how chances goes when you set average = 2.0 instead of 1.5 for same min=0 and max=3,

0 - 7,02%
1 - 10,12%
2 - 48,91%
3 - 33,95%

Now tests with classic min=1, max=6, average=3,5
1 - 10,01%
2 - 12,02%
3 - 27,58%
4 - 28,59%
5 - 11,83%
6 - 9,97%

So extreme values has 10% chance. Average value 3 and 4 have 56%.

Tests with min=2, max=5, average=3,5
2 - 14,82%
3 - 35,25%
4 - 35,15%
5 - 14,78%
15% chances to get extreme value

Tests with rather usual min=2, max=5, average=3
2 - 3321
3 - 4988
4 - 1032
5 - 659
6-7% percents to get 5 instead of 25% chances before. And even taking 4 or 5 is now much harder - only 17% totally.

So, i`m asking all GMs to rethink their generation params. At first you can set max 5 without doubts, really low chance to get nice leader wit that. At second classic scale, when you set Average to 33% between Min and Max (like 2 min, 3 - avg, 5 - max), gives only 17% chance to get over 3.
At the same time 50% receives average (3)

Maybe later the formula should be soften to increase extreme values.
 
btw, ForzaA, our discussed method didn`t work.

i did following with ready random value:
$value=($value-$middle)*($value-$middle)/(100-$middle)+$middle if $value >= $middle;
$value=$middle-($middle-$value)*($middle-$value)/$middle if $value < $middle;

$middle is defined average
 
Added: By request of DanielA you can set individual per each country average movement, fire and shock instead of global one.
 
Tonioz said:
Added: By request of DanielA you can set individual per each country average movement, fire and shock instead of global one.
Click to expand...
Damn good job that!!! :)
 
post in Test of Skill with explanations of some options of generator. Useful for all. Related with this generation: http://www.europa2.ru/cgi-bin/leagu...ason=Valkyrie.net VIII&lang=eng&yearsave=1434

1. Are you sure you want everyone would get 2 general and 2 admirals each session ? You can make number of leaders random too, for example from 1 to 3 or so. You can let people choose leadersgroup - LT choice like it was in past games like:
LT1: 3 generals, 1 admiral
LT2: 2 generals, 2 admiral
LT3: 1 general, 3 admirals
or any variations you wish.

2. Life period exp/conq. I guess you wanted to set it on 7-8 as average ?

3. Time period for leaders generation - that is linked with average length of the session. It is average time during which the leaders should generated if "Time of leaders appearance" is choosed "During the session".

4. Time of leaders appearance. If you set up "At start", they appear at first year of the session. During session will make appearance of the leaders random.

5. Life period gen/adm. It is 15,25,30 now. It`s up to you, i suppose 20 years average is all right.

6. Shift leaders appearance (years from start). Very rare option, asked by KJ once. If you want to delay leaders (for example you just had very small session), set this value like "5". So first leaders will appear not in 1434, but in 1439.

7. Show generated leaders - you can hide them from everybody including yourself by choosing "no".

8. Influence of land slider of number of generals/admirals
You can make number of leaders depending on land slider of each nation automatically. For example, set 10% dependance per click.

Imagine you want generate following number of leaders:
Generals:
min - 0
avg - 1
max - 2
Admirals:
min - 0
avg - 1
max - 2

Fully land nation (land=10) will get 50% bonus (5 clicks * 10%) of getting more generals and 50% malus of getting admirals.
So land=10 will get following number of leaders if you use 10% setting:
Generals:
min - 0
avg - 1,5
max - 3
Admirals:
min - 0
avg - 0,5
max - 1
 
Update:
This update is done to able to do better two different generations for the game:
- first you generate extra leaders for certain nations, like Venice 1419-1800. Then you upload save with leaders to stats, and generate for other country and required period etc
- at second you generate "usual" random leaders between session.

The problem that Generator used Gaussian distribution to make values are around center (average). But for generating leaders for many years in advance you need Linear distribution. For example for Venice 1419-1800 you need leaders were spreaded to whole the time equally, and not stack them around 1600.

So:
- "ID Range" added. Use 34000-34999 range to generate leaders for whole game. Use 35000-35999 range , as usual, for random session leaders. Just don`t cross this ID to prevent double ID of random leaders.
- The choice of Gauss/Linear method is added. So you can choose manually do you want equal chances/more values around average for the result of generation.
- The choice of different identification signs added instead of only "`"
 
not harder than usual controller of injection molding machine for plastic :rofl:
 
You must log in or register to reply here.
Top Bottom
Back