Rethink the limit of special classes

[harthag]

harthag I'm not sure if you are talking about setups with unlimited classes, but with unlimited classes you will likely see snipers all over, maybe sometimes shotguns for shotgun rush.

Yes, that was all based on the OP's suggestion of revised limits on special classes, though not necessarily their specifically suggested implementation.

I did not like 5.2. much. I would prefer the actual situation way more.

I wasn't here for 5.2, so I have no personal experience to draw on. The examples I gave were for illustrative purposes only.

Without changes to stats, I agree that the current situation is best. My main ppint is that if the current limitations are relaxed, then appropriate stat changes should necessarily accompany that change. With the secondary point being that those changes "could be" an improvement, if done well.

[Captain Spock (Castlevania)]

With all the parameters given, the assault is perfect in being so average. To make it more competitive, we should not change one parameter to make it more powerful, because it would destroy the nice balance. This game is designed to start with Assaults, built you AI for Assaults, then unlocking new bots and specialize them. That's good.

Are you serious? Please go to see the data before drawing such a conclusion:
http://forum.gladiabots.com/viewtopic.php?t=429&p=4659
http://gfx47.com/games/Gladiabots/Forum/viewtopic.php?f=8&t=17&p=4111&hilit=bot+stats+shield#p4111

We see that compared to shotgun and machinegun, assault excels in one attribute but fall behind in at least three. If the data above are not persuasive enough, I made two 1 vs 1 experiments:
1. For both, move to their preferred range and shoot. No retreat.
2. Both use your GrandMaster level Simplebot AI.
Here is the result:
1.
SG wins SN, AS; loses to MG
MG wins all.
SN wins AS, loses to SG, MG
AS loses to all.
2.
SG loses to all because it cannot damage from long range.
MG wins AS, SG; loses to SN
SN wins all.
AS wins to SG, loses to MG, SN.

However, if I change the program to forbid Shotgun shooting at long range, I get the expected circle: SG>SN>MG>SG
AND Assault loses to all.
[

]harthag I'm not sure if you are talking about setups with unlimited classes, but with unlimited classes you will likely see snipers all over, maybe sometimes shotguns for shotgun rush.

It is not wrong for being average, but it is bad that the general one cannot win a special one in a general case.

Although I didn't play at that time, I know it would have happened and why. Seeing the well balanced games like Dota, the special ones always have one strength but more weaknesses . That's why they need complementary teammates. Here we have the weaker ordinary force and strong special force, so not hard to imagine what would happen if special force even accumulates. The dominator is Sniper simply because it is the easiest to be programmed powerful. The ideal case for this game is that the players uses different classes because of the value of complementary instead of being forced to.

[PEW]

Are you serious? Please go to see the data before drawing such a conclusion:
http://forum.gladiabots.com/viewtopic.php?t=429&p=4659
http://gfx47.com/games/Gladiabots/Forum/viewtopic.php?f=8&t=17&p=4111&hilit=bot+stats+shield#p4111

Of corse I checked the data before. AS is average in short damage, average in medium damage and average in long range damage. It is also average in moving speed, has a weaker shild than MG and SG but higher health then SN and SG, which also makes it average. So in my conclusion the AS is average in every important parameter. That does not make the AS an average strong bot, because others can be specialized to use their benefits, while the AS has no strength besides being harder to count. In fact, the AS is the weakest bot, maybe I did not make that clear before.

We see that compared to shotgun and machinegun, assault excels in one attribute but fall behind in at least three. If the data above are not persuasive enough, I made two 1 vs 1 experiments:
1. For both, move to their preferred range and shoot. No retreat.
2. Both use your GrandMaster level Simplebot AI.
Here is the result:
1.
SG wins SN, AS; loses to MG
MG wins all.
SN wins AS, loses to SG, MG
AS loses to all.
2.
SG loses to all because it cannot damage from long range.
MG wins AS, SG; loses to SN
SN wins all.
AS wins to SG, loses to MG, SN.

However, if I change the program to forbid Shotgun shooting at long range, I get the expected circle: SG>SN>MG>SG
AND Assault loses to all.

Some days ago, I did a similar 1v1 testing with every bot combination, every useful range combination (where to stop pushing) and multiple strategies combination with a total nearly 50 different setups and wrote all of their results, successful behavior, achieved damaged received damage down for analyzation. The big difference is that I can now make general conclusions while you can't in some cases because both of your bots used the same AI in the two test runs.

So one different result I got is for example, that a smart AS definitely wins a 1v1 versus a SG if they both start at long distance, no matter what the SG tries to do.
Also, the SN is not that strong, losing always against a smart SG or MG even if they start in long range, no matter what the SN tries to do.
One of my results is, that the MG is the strongest 1v1 bot.

[TheKidPunisher]

There is a map where it is a 1v1? I do not know why complaining about which class is stronger in 1v1. Maps are massively teamplay between all bots. Assault are good for that, they can support other bot in many ways due to their versatility.

[GFX47]

What about a budget system?
Each class has a cost, special are obviously more expensive and each map has a max budget. We'll have to find the right balance of prices with time of course.
Don't forget there will be more classes in the future.

[pier4r]

I do think I budget system will do it, the problem is to determine the value of a class. Unless in the database you have the stats for every game, like damage delivered, damage taken, distance walked, resource taken, and so on. With a database like that we could extract an index of relative power.

Anyway if we would approximate like in chess (where pieces have "pawn value", that works in most cases) I would say.

Assault 1000 (reference)
MG 1400
Sniper 2000
Shotgun 2200

[Captain Spock (Castlevania)]

What about a budget system?
Each class has a cost, special are obviously more expensive and each map has a max budget. We'll have to find the right balance of prices with time of course.
Don't forget there will be more classes in the future.

The problem might be: If the prices are too rough, experienced player will still use the obviously better configuration; if the prices are made too accurate, the players would face a very tough knapsack problem.

Balancing stat would make it easy. Currently, I think all specials are will balanced, the problem is only the relationship between specials and assaults.

Chess actually gives us some good indications:
Bishops and Knights behave very different, it is literally hard to evaluate their accurate value, but in most cases, it is a good practice to value them the same.
1 bishop 1 knight 4 pawns vs 1 bishop 1 knight 4 pawns is interesting, 2 bishops 4 pawns vs 2 knights 4 pawns is also interesting, but 6 bishops vs 6 bishops or 6 bishops vs 6 knights is not interesting II know it because I used editor in computer chess program).

[sollniss]

I thought with random maps we get random classes as well?

What stands against deciding the classes randomly? Of course it would have to be mirrored but with that we kill off rock-paper-scissors completely and forever.

Say we have a pool of 5 bots for a map. Just decide randomly how many bots become what class. If your AI is really good it should win 5 Snipers vs 5 Snipers just as well at 4 Shotguns 1 Sniper vs 4 Shotguns 1 Sniper or 2A/1MG/1S/1SG vs 2A/1MG/1S/1SG.

That's the single fairest approach. And if you argue about configuration "skill", thats also irrelevant because on the current maps the configs are pretty static anyways and on random maps you won't see the map beforehand to adjust anyways.

Making it random would also spice up the current maps a lot more and give generic AIs more power which would make the game much healthier imo.

[mcompany]

With all the parameters given, the assault is perfect in being so average. To make it more competitive, we should not change one parameter to make it more powerful, because it would destroy the nice balance. This game is designed to start with Assaults, built you AI for Assaults, then unlocking new bots and specialize them. That's good.

I really can't agree with you on this one. At the very least, assaults are perfectly average, and that alone is a significant nerf to the bot class, or assaults are not average, but below average.
If we compared assaults to shotguns, assault gain more power in long range (which doesn't say much besides the fact that assaults can actually keep a shield down by itself) and in medium range (which is often negated by shotgun's speed), but they lose so much more. Shotgun has the same speed carrying resources, shotgun has a much faster move speed, shotgun's health+shield is the exact same as assaults health+shield when both bots a have no damage on it (and since more of shotgun's armor is in the shield, it is overall better), and shotgun does much more damage in short range (even more than machine gun's, which I'll get to).
If we compared assaults to machine guns, assaults gains in movement speed (which is definitely machine gun's biggest weakness), it has better accuracy in short range (unlike all other ranges where the accuracy is equal), and it shoots it's first bullets earlier (which rarely makes a difference). However, despite the accuracy bonus in short range, machine gun still does a good deal more damage in all ranges, and to top it off, machine gun has 2000 more shield to waste and 2000 more hp to waste which definitely helps overcome a good portion of it's slow speed
If we compared assaults to snipers, assaults have 2000 more hp and has the ability to gain damage just by moving forwards, but sniper's damage is comparable to any other bot class in medium range, so while it is slacking in short range (as it should be), it is ridiculous in long range (which is the range with the largest gap anyways and probably the most important range).

So, we could argue all we want about how shotgun loses to assaults in long range (as it should), it is still by far the weakest class with the most and largest of weaknesses compared to other bot types, which is why I say that assaults are highly underpowered and only makes since if there are a lot of them

As far as I know, diversified strategies is boosted by diversified configuration.
In my suggestion, the total kind of special classes will decrease in very small maps, but could you please illustrate a "strategy" that requires coordination between different special classes? At least I don't have one. What I see is just that you cannot use "dancing shotgun" and "shooting at the edge" snipers at the same time in Mind Game. Ironically, Mind Game is the most notorious rock-paper-scissor in current rule.

First off, Mind game is and has always been a map where the rock-paper-scissors of bot classes happens the least. In fact, the only times it had a rock-paper-scissors between bot classes was from 5.3 to 6.1, where most players was still trying to adapt new strategies for that map, either because we was adapting to not having the free ability of bot class like in 5.2 or because was adapting to playing with tags only. Even during 5.2, double machine gun pretty much became the standard, and the rock-paper-scissors became more between the strategies you chose or if you was willing to try to gamble on the enemy's AI and put a sniper in the back.
As far as other dynamics go, like pier said, 5.2 almost always came down to either full sniper teams (which keep a lot of power and was at least somewhat versatile) or full shotgun rushes (which was only not as good because people kept looking for ways for snipers to beat them and machine guns easily beat shotguns (although more machine gun AIs wasn't as developed at the time as they are now, so shotguns still could win)). I've done several tests back then to see if there was any diversity that was useful, and besides pushbot (which had to many weaknesses in my opinion to be viable and had (and still has) even less nodes to support it), there wasn't much luck. Machine guns are too slow to use well with shotguns or assaults (if you want to notice machine guns working with assaults, check out Kingmaker with sniper-assault-assault compared to sniper-machine gun-assault), and assaults have paper armor for their not so great movement speed (or even just in general), and sniper shotgun is just a gimmick that only works on abusing both the opponent's AI and the shotgun's speed. In general, most of the diversified teams in 5.2 was to avoid being hit by a rock-paper-scissors (which often times lead to what I complain as an "unbalanced layout" where one player might only win because of a gamble of the starting bot placement rather than just the strategy, which could be argued to happen all the time these days).
So, getting back to the original question (which I honestly forgot when I typed that paragraph), no, there are no diverse strategies that could be gained by having different bot types

[mcompany]

What about a budget system?
Each class has a cost, special are obviously more expensive and each map has a max budget. We'll have to find the right balance of prices with time of course.
Don't forget there will be more classes in the future.

I thought you was already thinking of how to do this, and I personally am hyped for having such a mechanic (which would definitely give this game much more of an RTS feel)


PS: I still think that pushbot should be allowed again

[Kanishka]

And while the others were discussing how to manage the new classes and balance the old ones, one little boy sat in the corner, despairing that the game is no longer focusing entirely on AI creation...

Why not do something like

Career Mode Classic: One Class. Entirely AI focused. Grants more XP or something.

Career Mode Zen (or something.): Classes, Customization, Perks, etc etc. Grants less or no XP.

[Kanishka]

If we keep adding more 'features', we'll end up with a 'Candy Crush', 'Temple Run' or 'Clash of Clans' clone. Talking about CoC, we may get "Actually, maintaining a Bot HQ with AIs assigned to the guard Bots isn't such a bad idea..."

[mcompany]

And while the others were discussing how to manage the new classes and balance the old ones, one little boy sat in the corner, despairing that the game is no longer focusing entirely on AI creation...

The problem is that the balance of bot classes is a big deal, especially when there is classes like the sniper and the shotgun. I don't think there is an automatic loss in the complexity in AIs once the issue of bot classes is resolved, but the overall balance of the game is dependent on bot classes, and the AI complexity depends on how well this balance is done

Career Mode Classic: One Class. Entirely AI focused. Grants more XP or something.

Career Mode Zen (or something.): Classes, Customization, Perks, etc etc. Grants less or no XP.

I'd be fine with the Classic mode, but after that, any other modes I'd say should be like a 5.2 mode, or the current 5.3 mode, or a budget system mode, or just a full on RTS game

If we keep adding more 'features', we'll end up with a 'Candy Crush', 'Temple Run' or 'Clash of Clans' clone. Talking about CoC, we may get "Actually, maintaining a Bot HQ with AIs assigned to the guard Bots isn't such a bad idea..."

What does either of those points mean exactly? On the second one, I'm pretty sure that plenty of people have tried to maintain plenty of "guard dog" AIs with snipers or machine guns

[pier4r]

Nice points mcompany.

I like the budget system, although it will require a bit of balance

Or completely random assignments (even with fixed map, but with random bot assignments. Working out in which position the bot is should be so much work to be practically impossible).

I mean even having fixed maps does not matter if the player cannot choose.
Imagine when you deploy a game that you select only the ai for the bot classes. Then the server chooses the map (not seen by the player, he can see the map after the deployment), the side, the bot classes (with current limitations or like 5.2) and their placement . This would effectively mean a lot of possible starting configurations that are unpractical to be detected (unless one takes Gladiabots as a job) , therefore even with a fixed map, one has to have good generic Ais to win.

Moreover the team would not surely have a sniper (or shotgun or whatever other class), because with random assignments one can end up (as well as the opponent) with just assaults. So one has to have really good generic Ais for all classes.

Sure, this makes testing even longer in terms of needed testing matches.

[GFX47]

I'd like to find a solution where you have a constant team (like in XCOM for instance).
I don't see any other way to make bot customization viable.

[Ritter Runkel]

I like the assault as they are. It's the only bot class you can have multiple times in your team. With this ability and the lack of counting capability, the assaults normally are the ones harming the other well ballanced bots.

I like the maps where you have some more assaults like in cod or spiral. It's really hard to decide when to retreat and when not. And that's only because there are several, uncountable assaults.

That's what makes them unique and imho well ballanced.

[Captain Spock (Castlevania)]

I like the assault as they are. It's the only bot class you can have multiple times in your team. With this ability and the lack of counting capability, the assaults normally are the ones harming the other well ballanced bots.

This is not necessarily true. One can count as many assaults as possible in current system. However, I count only 2 because I'm lazy

[mcompany]

I like the assault as they are. It's the only bot class you can have multiple times in your team. With this ability and the lack of counting capability, the assaults normally are the ones harming the other well ballanced bots.

This is not necessarily true. One can count as many assaults as possible in current system. However, I count only 2 because I'm lazy

It's possible but more tedious to figure out and more risky to count (assuming you are counting using tags. Any other method just wouldn't be the most accurate). Still, while I don't have a problem​ with assaults on maps like Barred Spiral or CoD (simply there are a lot of them), I think that the ability to have multiple of them is all that great of a reason to ignore their stats