Automatic Balance Tester - Page 3

On December 14 2013 02:38 gonzaw wrote:
Ideally there shouldn't be modkills.

Maybe more ideally, when someone is "modkilled" for inactivity, what actually happens is that the game is re-balanced (somehow) as if that player didn't join the game in the first place. Eg if there are 12 townies and 3 scum, and a townie is "modkilled" for inactivity, then you rebalance the game as if there were only 11 townies and 3 scum all along. The game doesn't "lose" balance (i.e if a townie got modkilled, town doens't lose advantage, if scum was modkilled, scum doesn't lose advantage).
How? Beats me, that's why it's idealized

On December 30 2013 14:06 wherebugsgo wrote:
Cool, I was looking toward making something like this myself. I just got time this break so I'm going to try to make something independently (partly cause I hate java) and I'll see what I get.

If you wanna collab on something let me know. I'm thinking of including past games and other types of data as well.

On December 30 2013 23:29 Xatalos wrote:
... Actually the logic of this program is pretty simple, and the hardest part has been to get the GUI working properly...

On December 30 2013 23:29 Xatalos wrote:


That sounds interesting. The approved setups here on TL Mafia have already gone through a manual balancing process, so they should be a good starting point for automatic balancing. Are you thinking of basing the whole balancing process on past games data or just showing them as some kind of a reference for the user? kitaman's TL Mafia Database would probably be helpful for both purposes.

Which language are you planning to use? How similar is it to my current balance tester? I could share the code I have currently on this one if you want. Actually the logic of this program is pretty simple, and the hardest part has been to get the GUI working properly (it still isn't working completely properly ), but it might still be helpful for you to glance at the logic portion. It might also be helpful for me if you find something to improve.

Depending on how similar our aims are, it might make sense to maintain two separate programs. If they're very similar, though, it might not make much sense. Currently my main goal is just to make the program easy to use and fairly reliable. I thought of using iGrok's balancing method for a bit, but ultimately I didn't know how to automate it. The +/- method is very simple, but its accuracy suffers in larger / more complicated setups. The best option would probably be to calculate every possible scenario each day/night for the whole game, but that gets very complicated very fast. How were you thinking of calculating the balance?

On January 01 2014 07:37 wherebugsgo wrote:


Probably at least partially based on past setups, though the sample size is incredibly small.

More likely what I'll be doing is encouraging the people on the balance team to run the setup randomizer I create lots of times. I'll probably put in some sort of rudimentary learning algorithm that will assign a balance rating, i.e. + for town, - for mafia similar to what you have. Based on what the tester inputs as his opinion the program will change its future predictions for the setups it creates/sees.

It'd hopefully become more effective as more games are added. I'll likely also include multiple options for balance calculation. I'm thinking the main ones would be:

1. Testers' inputted scores
2. Role balance based on game winrate when included
3. Balance based upon +/- scores given to different roles (manually, not through past games)

So there'd probably be a bunch of checkboxes or input boxes for each of the calculations, and the program would just keep track of a variety of different values to calculate scores to maintain versatility.

In other words, if I wanted to see what the program thought without including past games, there'd be an option to turn off #2. If I wanted to manually tune the score based on what I thought certain roles are worth, I could change scores on the fly for #3.

e: unfortunately I don't have a proper coding platform atm, I fucking hate Windows. So this'll probably be on the backburner for a bit.

also I'll probably write it in either Python or C++.

On January 01 2014 10:29 gonzaw wrote:
I think this would probably work better as web app. You could use a free cloud hosting system perhaps?

Also, how about making this open source? Msybe open a github proyect so people can cooperate

On January 01 2014 17:10 Xatalos wrote:


Yeah, I think a web app would be more convenient. It's just that I have basically no experience with web apps so far... It's definitely a possibility for the future, though.

I've also been thinking of making the program open source. It's not like I'm planning to make money off of it or to maintain it forever. However, I think there's still some work to be done before I'm ready to actually "launch" the program

On January 10 2014 10:54 gonzaw wrote:


Well....ok

With a web app, you'd have less "control" over the GUI (like you'd do with a desktop app), but you have more powerful frameworks to choose from to help you. E.g databases, hosting services, bla bla bla. Well, if you choose specific frameworks (which you have to research), and you know lots of javascript, or html5, etc, you can have a pretty bitching GUI. However that's harder than just creating an app with Swing. Although I think JavaFX may be better perhaps; I never really liked Swing.

Also you already have it hosted, nobody needs to install JRE or download the app or anything. Because it's hosted in a single place, you can have users, and statistics, and shit like that



Yeah, it's kind of a "tough" thing to do. If you are the project leader, you'll have to do lots of stuff, and basically learn a whole new industry setting and procedures, etc.
For instance here's a guide: http://producingoss.com/en/index.html
It does take a while to learn, and then get enough experience and practice and collaboration and shit (not that I ever did any open source project though lol).

If you do go open source though, you can gather the power from anybody ever that's relatively interested in tools to play mafia, instead of just doing it yourself. That has the potential of creating a super awesome tool, and hell maybe even advancing the "game theory" behind mafia (regarding balancing in this specific case for instance).


I'd volunteer and stuff.....but I'm lazy (you can notice I haven't done anything on that other project I wanted to do and made a thread about like one year ago lol).


Also, as for suggestions, I would suggest:
-Include a database of the games it's being used on. Whenever the user chooses to balance a game, you can ask for the name of the game and that kind of stuff, and save it for later (for example, to later check how that previous game was balanced once it's over).

-Migrate kita's database of games into this new database of yours. That way we can see how every single game from TL is balanced....and well that's nice. It's not hard work as long as you create a compatible infrastructure and create a script to migrate kita's database

On January 11 2014 04:38 kitaman27 wrote:
I know this is kinda outside the realm of what you have currently, but something that I find really helpful is knowing the cycle the game would end in worst case scenario for each faction.

For example, 2 Goons/7 VT would end D3 at the earliest for a mafia win or D2 at the earliest for a town win.
2 Goons/6 VT 1 Vig would end D2 at the earliest for a mafia win or N1 at the earliest for a town win.

If you're looking for a feature to add at any point, it would probably be fairly simple to come up with a formula for this for mini games.