Re: Balance Command

**blu.knight** · 05-31-2007, 09:35 PM

How do you implement nueral nets/what are they good for? It's the only part of CS that seems kind of like magic to me... and it's unlikely I'll be taking a class on it.

**Stupid** · 06-01-2007, 12:19 AM

Wiki is your friend: http://en.wikipedia.org/wiki/Artificial_neural_network

**Yer Mom** · 06-02-2007, 03:24 AM

I think the most obvious concern with the plugin is the fact that it sometimes assigns the majority of the "pro" players to one team. At least, that's my concern.

The plugin itself seems to be working fine short of that fact. Every few games that are balanced seemed to put ALL the "pro" players on one team, and all the mediocre players on another. I'm wondering if it's a glitch, but I've been told the plugin is to be working "properly". That makes me question what the plugins intentions truly are.

Can we not just assign a * by "pro" players in the data and make the algorithm sort them between teams? Maybe a * system where the better the player is the more stars they have? I don't really know how the whole system works but I'm pretty sure that it works by steam ID and that it's probably possible to tag certain steam ID's so that they are treated differently by the plugin. Maybe we could even collect data from a plugin that tracks individual statistics. Hit percent, damage per game, kills per game, most kills, most deaths. I'm not asking for that information to be public as I've seen what it can do to a server, but I am wondering if it would be useful for the admin to track this statistical data so that they can assign individual *s (or however that works) to each player. Then make the algorithm assign teams by going down the list. One 5 * player Marines, one 5 * player Aliens. One 4 star player Marines, etc., etc., etc.,.

Maybe all that's not do-able. I dunno since I don't code. But I have done PLC coding before and I know that if these algorithms are anything like Ladder Logic then it is possible as statistics make up 90% of PLC programming.

Although, those are two different worlds, lol.

-Mom

**Kerostasis** · 06-02-2007, 12:08 PM

The view of which players are "pro" and which players are not is entirely subjective, and the plugin has no idea which players you personally think are pro. It just tracks win records. And it must be doing a fairly good job, because only a tiny number of people on the server have maintained a win record more than 10% away from 50/50. When the plugin is sorting between 4 players whose records are 58%, 57%, 57%, 55%, its entirely possible to reach a configuration that subjectively looks stacked when objectively the numbers are just about the same on each team. Using 5 *s instead of win % wouldn't help much, since most peoples win ratios are already so close together.

**Agamemnon** · 06-02-2007, 04:12 PM

So there are two things that cause the games where all the 'pro' regulars are on one side. The most common is that a relatively new player has recently reached the ten game threshold and has either a very low or very high win ratio, which causes the plugin to try to balance that by putting the rest of the good players against him (or with him if he's really low). It was known that this would occur, but the server seems to be having many more new players than we expected, which is making this more frequent than I expected. Currently we start everyone's record with 1 win and 1 loss to avoid extreme values, I'd like to change that to 10 wins and 10 losses, alternating so they don't cause a problem as they drop off the end. This should minimize this problem if we can get this change made.

The other thing that can cause this is that there are some players who do have a low win ratio, and the plugin takes them into account just as much as the players with a high win ratio. So what may occur is that one team will have 4 'pro' players and 4 'less skilled' players, while the other team has 8 solid middle of the road players. The team with 8 solid players will then think that the teams are stacked against them, after which they may well win and wonder how they managed it. The only time this becomes a problem in my view is when a player is keeping their ratio artificially high or low by repeatedly leaving in the middle of a game (this is usually unintentional, but a player who has a tendency to rage will become overvalued, since their losses are often unrecorded). Anytime that player is on the server, the teams may be uneven. I don't know if it's possible to assign a loss whenever someone leaves in the middle of a game, nor am I sure that's a good idea even if it is possible.

**1Stop|Beowulf** · 06-02-2007, 05:29 PM

the only other problem i can see with balance is that the better it does, the less effective it becomes. it's goal is to even the teams by giving everyone a 1:1 W:L ratio, but the closer each player gets to a 1:1 ratio the less effective it is at balancing said player. so maybe now we're just feeling the result of the high end of the balance plugin's cyclic effectiveness.

**blu.knight** · 06-02-2007, 06:16 PM

It does not take everything in to account. Win/Loss is not equal to skill level. If you get stuck with people that are very bad with teamwork, you will probably lose regardless of your skill. On the other hand, if you're not very good, but you happen to more frequently be on a team with people that work well together, your ratio will raise, and the balance plugin will be trying to split you up from the players that win more... which might end up hurting the team you go to because your skill level could be far less than the others.

I said it the last time this topic was around, balancing on win/loss alone in a team game does not work. It doesn't take into account individual skill nor an individual's leadership ability or effectiveness in a group.

**Agamemnon** · 06-03-2007, 02:02 AM

Blu, everything you mention there effectively functions as random noise, which will be insignificant for the large sample size we're using, unless there's a systematic bias. Convince me that something causes you to repeatedly be put on good (or bad)teams and you'll have a case for a systematic bias. Balancing based on win/loss by definition takes into account everything that affects winning or losing. Anything that causes you to win more or less frequently will affect your W/L ratio, and hence shows up in the data.

I agree that there are some biases that we can do nothing about, such as frequent mid-game disconnects or the inherent bias of playing at different times. Any other systematic biases should be dealt with on a case by case basis if possible. However, I assure you that with a large sample size the random noise you're describing will not have any significant effect. Only players that have a small sample size (new players) will be affected by that type of noise.

**Yer Mom** · 06-04-2007, 12:09 PM

Originally Posted by Agamemnon

The other thing that can cause this is that there are some players who do have a low win ratio, and the plugin takes them into account just as much as the players with a high win ratio. So what may occur is that one team will have 4 'pro' players and 4 'less skilled' players, while the other team has 8 solid middle of the road players. The team with 8 solid players will then think that the teams are stacked against them, after which they may well win and wonder how they managed it. The only time this becomes a problem in my view is when a player is keeping their ratio artificially high or low by repeatedly leaving in the middle of a game (this is usually unintentional, but a player who has a tendency to rage will become overvalued, since their losses are often unrecorded). Anytime that player is on the server, the teams may be uneven. I don't know if it's possible to assign a loss whenever someone leaves in the middle of a game, nor am I sure that's a good idea even if it is possible.

This statement was proven true just recently, which is why I haven't returned to this thread. That game was outstanding.

It was 8 entirely middle of the road players on Aliens, and 4-4 pro/less pro players on Marines. The teams looked ENTIRELY stacked as we had no names such as Dirm, or Civilian and they had all of them. As the game progressed, we held ground easily due to teamwork and response time. I started to realize that these pro players suck when they have 4 skulks and a lerk biting at their heads.

I think we may be heading in the wrong direction with this. Although, I'm probably wrong. Your implementation is working well enough for me, but I'm concerned that if teams get balanced this way much more in the future, we may have problems from players. Teamwork outbalances skill any day. We may need to take that in to account when considering the balance command. Over stacking all the teamwork oriented players on one team may prove to be just as frustrating as stacking all the "pro" players on one team.

I'm not saying it needs to be changed, I'm saying that that needs to be taken into consideration.

-Mom

**Stupid** · 06-04-2007, 03:02 PM

You can't measure everything you need to have a "perfect" balance every map given 16 "random" players. Even by hand you can't do it. Even the best of captains can choose lousy teams. Heck - a player can be having an off night that throws everything about their past performance out the window.

What we've got is working better than amx_random or waiting for people to join themselves. Even when I feel like the opposing team (from me) is super-stacked, I still have a good and enjoyable game. That's all I can ask of the plugin and it works much more often than it doesn't.

**blu.knight** · 06-04-2007, 05:52 PM

Originally Posted by Stupid

You can't measure everything you need to have a "perfect" balance every map given 16 "random" players. Even by hand you can't do it. Even the best of captains can choose lousy teams. Heck - a player can be having an off night that throws everything about their past performance out the window.

What we've got is working better than amx_random or waiting for people to join themselves. Even when I feel like the opposing team (from me) is super-stacked, I still have a good and enjoyable game. That's all I can ask of the plugin and it works much more often than it doesn't.

When there are two good captains up, the teams are normally even and the game is long and good. Humans are better at this than a machine. We actually can hand pick teams better than the computer can. We see teamwork and we can see which people work together. We see who can make a good team.

The people that are good at picking teams can see this. There is also more to captains picking because the entire picking of teams is a game in and of itself. The captain has to predict who the other captain will value. If said captain wants the same person they know the other one will want, they choose that one ebfore someone else who they may value more... but know the other captain won't pick.

I've said it before, I love captains. The team picking is fun.

**FireFly** · 06-05-2007, 03:27 AM

Originally Posted by blu.knight

I've said it before, I love captains. The team picking is fun.

Yeah sez you who isn't the fat kid who gets picked last and then told to be a fat perma gorge! *cries in the corner and eats a bag of cheesies*

**Stupid** · 06-05-2007, 07:18 AM

Originally Posted by blu.knight

When there are two good captains up, the teams are normally even and the game is long and good....

I said "can choose lousy teams." I agree with your "normally" qualifier and the rest of your comments. My point is that the balance command can only ever be a proxy for a (rarely) flawed human-picked game. The degree of success of the program is 1) its success in giving (mostly) good, long, even games, 2) the enjoyment of the average of all games, 3) and its ease and frequency of use.

I can't find any fault with the current implementation as 1) most games (to me) are pretty even, 2) I have far more fun games than not, and 3) it gets used pretty much every map.

Thread Tools

Display