Page 314 of 323

Posted: Tue Jul 02, 2019 4:52 am
by chennisden
found it

google

mafia game expected value

should be by erlin yao.

Posted: Tue Jul 02, 2019 4:54 am
by RadiantCowbells
I can't start arguing for changes without first establishing that they're necessary :P

Posted: Tue Jul 02, 2019 4:57 am
by Ether
I've seen the square root thing before, but it doesn't seem to make any sense in practice.

Even assuming that towns will do better than random isn't a reliable thing. Like, the newbie setup used to have a vanilla subsetup back in the F11 days, and the win rates there were just exactly what the EV would predict. Towns do much better than random in drawn-out Nightless games, but a lot of other mafia types MIGHT ACTUALLY BE just random bullsh--

Oh, hi, Raybells! Congratulations on your 75% town win rate!

Posted: Tue Jul 02, 2019 4:58 am
by chennisden
alright so lets try this

i actually think the ev isnt hard to find.

lets just pretend that the game keeps going after mafia all dies. this is okay because we define winning as:

when theres 3 people and the last guy gets lynched. there are TWO PEOPLE left. if BOTH OF THEM ARE TOWN, TOWN WINS

Posted: Tue Jul 02, 2019 4:58 am
by chennisden
ill talk about this tmrw need to sleep.

Posted: Tue Jul 02, 2019 5:03 am
by Auro
Are you saying that problem is equivalent to something like, "Pick two random from 11+4 and find the probability of both being town"? That would be wrong. I haven't seen any easy reduction yet.

Posted: Tue Jul 02, 2019 5:05 am
by Something_Smart
You should start a new thread for it, actually, since it doesn't really relate to this game.

I don't really feel like attempting to follow the math in that paper, but I can tell you I just simulated a million 13:4 mountainous games and town won 15% of them. There is no way that the EV is anywhere near 50%, regardless of what that guy's complicated math says.

Posted: Tue Jul 02, 2019 5:07 am
by Something_Smart
In post 7830, Auro wrote:Are you saying that problem is equivalent to something like, "Pick two random from 11+4 and find the probability of both being town"? That would be wrong. I haven't seen any easy reduction yet.
I think he's getting at a reduction where you randomly rank all the players and then town wins if X town are ranked higher than Y scum (more complicated than that, but that's the gist). That may work and if so would reduce EV to a single combinatorics problem instead of a shitton of probability problems.

Posted: Tue Jul 02, 2019 5:08 am
by RadiantCowbells
That view is the correct one

The one where you rank the players from top to bottom

It misses a lot but

Posted: Tue Jul 02, 2019 5:08 am
by Krazy
In post 7775, RadiantCowbells wrote:I think right now it would be won by town a lot because towns are very strong these days
LAL VOTE: rc

Posted: Tue Jul 02, 2019 5:10 am
by Something_Smart
In theory, if chennis's assumption is valid which it seems to be, you could use combinatorics to count the number of possible games and count how many of them are won by town.

Pedit: well yeah it's just an EV calculation. It's not meant to be a model of how players actually play, although as a very rough model it isn't the worst one.

Posted: Tue Jul 02, 2019 5:11 am
by RadiantCowbells
It's the one I used for my normal game WR calculator

Posted: Tue Jul 02, 2019 5:13 am
by Auro
Something_Smart, yeah, that's what I was implying he was trying to do. I was trying to reduce Nightless games into that kind of combinatorics problem earlier but didn't find one, doubt it's easy for this too.

Posted: Tue Jul 02, 2019 5:14 am
by Auro
Yeah, even when you rank players that way there are extra conditions to impose and that removes the "simple combinatorics" solutions you could get. Hmm, I'll try again though.

Posted: Tue Jul 02, 2019 5:15 am
by RadiantCowbells
Not that I agree with his statements about win equity

I just think the system of modelling is superior and it also works well with calculating the impact of cops and such

Posted: Tue Jul 02, 2019 5:19 am
by Auro
Would you be willing to share it at some point?

Posted: Tue Jul 02, 2019 5:31 am
by Nancy Drew 39
In post 7816, chennisden wrote:heres a question

whens the last time anyone complained about a setup being scumsided
Lots of games are. I know I’ve complained about a few.

Posted: Tue Jul 02, 2019 5:32 am
by RadiantCowbells
It's outdated it was based off of the old normal rules

Posted: Tue Jul 02, 2019 5:34 am
by Auro
I'm still interested though!

Posted: Tue Jul 02, 2019 6:53 am
by Creature
In post 2, BNL wrote:
Lynch
- Okay
1of4
- Should be disabled if there's only one scum left
8-ball
- Okay, but maybe make it easier for town to hit the 8ball?
Arena
- Okay
Call a Friend
- Can only call one player and if that player doesn't give anything no message will be displayed
Coalition
- Allow scum to nightkill
Dance
- After dance, undo dance pairs and allow scum to kill
Popcorn
- Okay
Protect
- Okay I guess
Quiz
- Have no idea tbh
Rank
- Okay
Reveal
- Okay
Treestump
- Okay
Vengeful
- Should allow scum to nightkill if town is the one to venge
Veto
- idk
Also thought about a mafia 1-shot repicker

Posted: Tue Jul 02, 2019 6:54 am
by Creature
This game was annoying because I was like the 7th player to die and it took about 7 game days for me to die. This game like is dragged way too much which is pretty time consuming for both sides.

Posted: Tue Jul 02, 2019 6:54 am
by Creature
Should probably disallow many mechanisms preventing any death to occur

Posted: Tue Jul 02, 2019 6:56 am
by Skygazer
8-ball - Okay, but maybe make it easier for town to hit the 8ball?
pick three players and all of them are two and two thirds balls

Posted: Tue Jul 02, 2019 7:00 am
by Dannflor
why not pick 17 players and all of them are just under half a ball

Posted: Tue Jul 02, 2019 7:01 am
by RadiantCowbells
why not pick 8 players and they can all have 1 ball