59 replies to this topic

[Khobai]

And the point of all this math is?

Do you have suggestions for how to balance weapons better based on your maths?

[MadTulip]

p4r4g0n, on 01 September 2014 - 03:21 PM, said:

I liked the analyses however it would have been interesting to see if the different classes of weaponry are capable of being balanced against each other rather than just the balance within the weapon class. Having said that, will there be further analyses on the projectile weapon class with similar data as the beam weapon analysis?

Balanced between different weapon groups: Yes. That is the 2nd step though. The nice thing about a single weapon group is that the unknown parameters affecting the individual weapons are minimal. Setting all normal laser burn durations to the same value and they would be directly compareable in this way.

In between weapon groups we can eigther find an explicit expression for all unknown variables, OR and that seems easier.: we just balance a single variable that defines the relation between two weapon groups. This variable can then be balanced by just fiddling around with it, or even better by using game results.

simple example for using game results:
if the team with only lasers always wins against the team with only ACs, the lasers are better. There are ways that allow to condense this information down to a single number of how much better the lasers currently are even if all attending mechs have a mixture of those weapons equipped. BUT, you need the results of many (~10.000) games (who won? which equipment did they use and so on) which i dont have access to.

So first: inside the group as they behave similar. Then reduce complexity so much that groups among eachother can be compared by as few variables as possible. Then use statistics on game results to find balances values for those few variables.

[MadTulip]

Khobai, on 02 September 2014 - 11:19 AM, said:

And the point of all this math is?

Do you have suggestions for how to balance weapons better based on your maths?

Yea. You mean i should put numbers. I did put relations between weapons inside one group so far. So clan normal laser can be balanced among eachother as can IS normal lasers. It is not clear yet how much stronger clan normal lasers should be vs. IS normal lasers. But that problem is located one level higher so to speak. Im working in that direction.

Putting just numbers for individual stats instead of working on the relations between them is exactly what has been done with balance over the last years. My main point is, that the system has to be desinged an that system will then spit out the numbers. You have to understand the engine and then turn a screw instead of turning whatever when you have the impression that the engine is not performing.

[poohead]

Interesting stuff. You can use pure weapon specification data to work out if a weapon is better or worse within its competitive set.

But the only way to truly validate this is if you can correlate wins with the presence of that weapon (e.g. in matches with all lasers, if all the teams that had more of a particular laser weapon equipped won more times, and that agreed with where that weapon was deviating from the 'best fit' line positively). If you had truly granular data like that and found a specific correlation there's a real case for using it as a guide to balancing.

Also using this method you could balance everything simultaneously 'up' or 'down', leaving the relative differences intact and balanced but managing to increase or decrease time to kill.

And it's science maths, not a random nerf gun!

[MadTulip]

poohead, on 02 September 2014 - 11:43 AM, said:

... correlate wins .

Yea thats what i tried to say two post above. They got the data though.

[poohead]

The tricky part though, is deciding where the line should be in the first place. Correcting the outliers is simple compared to that. What are your thoughts on this?

For example it could be balanced based on match times, or win/loss ratio not going too far from 1 (based on standard deviations perhaps) when equipped above the average number in an average match.

So many options, but they all need data!

[MadTulip]

poohead, on 02 September 2014 - 11:55 AM, said:

The tricky part though, is deciding where the line should be in the first place. Correcting the outliers is simple compared to that. What are your thoughts on this?

For example it could be balanced based on match times, or win/loss ratio not going too far from 1 (based on standard deviations perhaps) when equipped above the average number in an average match.

So many options, but they all need data!

The first line is a gameplay descision imo. I would start with the average of one existing IS weapon group as thats the most mature. I.e. normal lasers. Bring them on theire line, then start to scale pulse lasers from that. With ACs i currently see a lot of trouble. It will also be difficult to access the pinpoint weapons and theire relation to "stream" weapons as well as always hitting streaks, totaly spreading lrms and such. In these cases there needs to be the "pinpoint vs stream" relation defined ONCE so we can connect those two systems. This variable might then be balanced by correlating with wins or even be up for discussion or a gameplay desicion if this should be a sniper game or a brawling game. It not even clear how that value is connected to skill of the player. Somebody who sprays and prays anyway will underperform with theese while a skilled sniper will overperform.

Also it became clear by the effect of tons and slots on the efficiency that light mechs will have other weapons beeing effective then assault (they prefer the bulk light ones). this might average out over the same number of lights and assaults in one team though. Clans should in general be completly scaled from the IS structure i believe.

Its work in progress. Easiest things like bringing the same group in line first, next easiest things next. At some point 95% or 99% accuracy dont matter anyway as the human observer wont perceive it.

Edited by MadTulip, 02 September 2014 - 12:11 PM.

[DAYLEET]

That tulip is mad, MAAAAAAD i tell you.

[MadTulip]

DAYLEET, on 02 September 2014 - 12:10 PM, said:

That tulip is mad, MAAAAAAD i tell you.

what do you exspect after 2 years ?

[MadTulip]

Update to the OP
- There was an error in the UAC calculation. I assumed double tapping damage as twice the normal damage, where it should scaling with the possibility to jam instead. Plots updated accordingly. Now the ACs dont look that overly powerfull as they did yesterday but are more or less in the same area as the respective laser weapons which seems realistic from a gameplay experience perspective - the model gets better .
- I included a model for the required ammo instead of assigning a fixed tonnage of 2t per AC. Current estimate for required ammo tonnage for CUAC 2,5,10,20 is i.e. [1.4 2.0 2.9 4.3] tonns per match. Its maybe a bit on the low side -> see OP ammo section.
- current code in OP is updated to V2 accordingly

Edited by MadTulip, 03 September 2014 - 06:27 AM.

[Dolph Hoskins]

Just to be clear the numbers that make your weapon charts are based off of current in game weapon specs?

I have been following this thread with interest, but I just want to make sure what these projections are based off of since there are currently so many #'s floating about right now given the test server and people's personal tuning suggestions.

[MadTulip]

The Ripper13, on 03 September 2014 - 10:47 AM, said:

Just to be clear the numbers that make your weapon charts are based off of current in game weapon specs?

I have been following this thread with interest, but I just want to make sure what these projections are based off of since there are currently so many #'s floating about right now given the test server and people's personal tuning suggestions.

Yea its just game data. I try to find a model (a set of equations/ratios) to make weapons compareable and thus balanceable. Those equations are like DPS = damage/cooldown where damage and cooldown is data from the game and DPS thus only depends on data from the game.

I try to work from solid ratios like DPS to those rations which you can not find unless you make assumptions and try to keep those assumptions or free parameters to a minimum. I assumed i.e. that you need an average of 2.5t of CUAC5 ammo per CUAC5 per match to estimate the total tons and slots consumed when using one CUAC5. I also assume you bring an XL engine between 250 and 400, no ES, no FF, DHS and max armor. Yea i think thats all i added. You can calculate the same stuff for other settings (i.e. normal heatsinks). It just doesnt make sense yet to plot all possible parameter combinations which wouldnt differ that much anyway. Its more about finding the model or general structure of the system then giving a single number to be set from 1.8 to 1.75 here. In fact the main point is, that the ratios between the weapons instead of single values should be known so that we can derive i.e. the whole clan weapon set from the IS weapons by just making them i.e 30% stronger.

Edited by MadTulip, 03 September 2014 - 11:13 AM.

[Dolph Hoskins]

MadTulip, on 03 September 2014 - 11:09 AM, said:

I try to work from solid ratios like DPS to those rations which you can not find unless you make assumptions and try to keep those assumptions or free parameters to a minimum. I assumed i.e. that you need an average of 2.5t of CUAC5 ammo per CUAC5 per match to estimate the total tons and slots consumed when using one CUAC5. I also assume you bring an XL engine between 250 and 400, no ES, no FF, DHS and max armor. Yea i think thats all i added.

Ok thank you. I thought you did a nice job representing everything that could have a definite numerical value applied to it. At first I considered some of the info faulty due to the lack of "applications" value, like various tactical pros and cons, but then I realized that is not a factor that could work in any way with the info you do have.

Although the way you displayed the overlapping ranges did a nice job of showing basic tactical application.

By applications I mean how effective weapons can be in areas like rapid movement, fire and forget, traceable firing source, exposure time, and those kinds of aspects.

[MadTulip]

The Ripper13, on 03 September 2014 - 11:24 AM, said:

Ok thank you. I thought you did a nice job representing everything that could have a definite numerical value applied to it. At first I considered some of the info faulty due to the lack of "applications" value, like various tactical pros and cons, but then I realized that is not a factor that could work in any way with the info you do have.

Although the way you displayed the overlapping ranges did a nice job of showing basic tactical application.

By applications I mean how effective weapons can be in areas like rapid movement, fire and forget, traceable firing source, exposure time, and those kinds of aspects.

Yes the exact efficiency for an near infinity number of gameplay aspects cant not be accessed without statistics based on actual game results which are only available to PGI. It is however valid to assume that weapons inside the same group, like normal IS lasers do perform quite comparable if it comes to i.e."rapid movement, fire and forget, traceable firing source, exposure time," which is why you can argue that these impacts kinda "cancel out" as they nearly dont differ between weapons of the same group. Theire values should therefor be balanced such that they coincide with the same "line". You can then also conclude that i.e. Clan normal lasers should be "by a to be decided percentage" more effective on all ranges compared to IS normal lasers and by themselfs also on a straight line.

Your point is very valid for the comparison of i.e. ACs and normal lasers which behave differently starting even at "traceable firing source" and the direct comparison is thus not so easy. Still, also the ACs should be in line among themselfs. And the AC line should jsut differ from the normal laser line by one value which is the relative effectiveness between them. I.e. ACs might be balanced to be 5% less effective then lasers because they have the "traceable firing source" advantage. So after appliing my model there is still things to be balanced, it is just that the number of parameters is vastly reduced.

This results in staticstics of games won in which a certain number of weapon types were brought to the match will need far less examples (games observed) for the statistics to figure out what the exact impact of such unknown quantities like "traceable firing source" realy is. For the statistics id doesnt even matter if you can name the exact differences between lasers and ACs. It would just tell you after say 1000 games in which lasers where on the same line as ACs inside theire groups and among eachother that bringing ACs on average increases your win chance by lets say 5.25%. Then you known that the ACs should be 5.25% less effective then the lasers without knowing to what extend this is caused by them beeing less traceable or them scaring the enemy more due to theire sound.

Edited by MadTulip, 04 September 2014 - 03:14 AM.

[Garandos]

Something to consider:

If you Balance everything straight down on numbers, you run into the following fatal problem:

Gameplay becomes a question of math, not of choice anymore.

Take Starcraft for example, its about as balanced as you can get to balance, but is it still a game of Strategy? Hardly, its a game of mathematical solutions and executions
of said solutions, nothing more.

Applying this "perfect balance" to MWO would lead to an total decline for certain weapons, as they are, on a pure mathematicla point of view, weaker then others, providing a "perfect" Solution taking away choice.

But if you not only keep, but even work on a perfect imbalance (which does not mean one Thing beats everything all the time, but, one thing is better then another thing, in certain situations)

For ex. the C-UAC Weapon line follows a simple rule as it is, the larger the weapon, the more deadly it becomes, not only in single fire, but in double tap aswell.

If you now normalise the dps gain per double tap, the CUAC 20 becomes weaker then the CUAC2 because i have the drawback of the short range on that weapon, meaning, i can put out more damage using the CUAC2 while you are still closing the gap, then you can after closing the gap, with using the CUAC20.

Adding in weight, i can even take 2 CUAC2´s for your one CUAC20 and by chance, simply kill you long before you can put that 20 to use.

Maybe a sim would be in order, to check, on the current state of weapon, if 2 mechs engage from max range, Firing weapons and hitting each time, while one is closing in at max speed, how much damage each would be able to deal under current stats and under proposed stats?

[Tombstoner]

You’re never going to balance DOT vs. FLD without a precision cofactor. Well that is you can get to a spot where you can call it balanced much like PGI has done, since the "feel" of the game then become the primary factor for judging balance. so as long as it feels good the math doesn't mean much.

[MadTulip]

Garandos, on 04 September 2014 - 03:19 AM, said:

Gameplay becomes a question of math, not of choice anymore.

Thats life.

Garandos, on 04 September 2014 - 03:19 AM, said:

Applying this "perfect balance" to MWO would lead to an total decline for certain weapons, as they are, on a pure mathematicla point of view, weaker then others, providing a "perfect" Solution taking away choice.

It is the opposite. There is only one perfect solution if things are out of balance. If things are in balances every element is equaly strong and it doesnt matter what layout you bring.

Garandos, on 04 September 2014 - 03:19 AM, said:

But if you not only keep, but even work on a perfect imbalance ...

Im showing what is imbalanced to what extend so that it can be nerfed by exactly that ammount to be balanced afterwards.

Garandos, on 04 September 2014 - 03:19 AM, said:

For ex. the C-UAC Weapon line follows a simple rule as it is, the larger the weapon, the more deadly it becomes, not only in single fire, but in double tap aswell.
If you now normalise the dps gain per double tap, the CUAC 20 becomes weaker then the CUAC2 because i have the drawback of the short range on that weapon, meaning, i can put out more damage using the CUAC2 while you are still closing the gap, then you can after closing the gap, with using the CUAC20.

This is why efficiencies are plotted over range. Youll notice that longer range weapons have less efficiency then shorter range weapons. This counters the problem you described. The exact falloff of damage over range is a single variable to be balanced which alows to shift from brawling to sniping metagame. There is an optimum steepness which rendering sniping and brawling equally efficient.

Garandos, on 04 September 2014 - 03:19 AM, said:

Adding in weight, i can even take 2 CUAC2´s for your one CUAC20 and by chance, simply kill you long before you can put that 20 to use.

You dont need to think in eigther damage, heat, range, tons, slots or ammo consumed alone as i already gave the model which is combining all those to a single value so that it is comparable. if this single value, the efficiency is bigger then the weapon is better. That is the point of the model.

Tombstoner, on 04 September 2014 - 03:37 AM, said:

You’re never going to balance DOT vs. FLD without a precision cofactor. Well that is you can get to a spot where you can call it balanced much like PGI has done, since the "feel" of the game then become the primary factor for judging balance. so as long as it feels good the math doesn't mean much.

It constantly doesnt feel good though. All the things you can not express explicitly are to be evaluated by statistics using game results as data. Whenever you try to crack an unknown system using statistics you should beforehand give as many known relations as you can. Then you should approximate in order to reduce variables where you can assume that your approximation doesnt include errors of relevant sizes. THEN is the point where you hammer the rest flat with statistics. Afterwards you can go back and release some of the approximations you did in order to compare different models for theire quality. Statistics wont tell you the exact solution. But they can tell you to what degree they are off. They also can be shown to be the optimal solution possible based on the data available while still not beeing accurate. I vote for choosing the optimal possible yet still error prone to a known extend solution instead of neglecting and doing whatever comes to mind.

Edited by MadTulip, 04 September 2014 - 03:47 AM.

[Iron Riding Cowboy]

Edited by Iron Riding Cowboy, 04 September 2014 - 03:56 AM.

[Ryvucz]

Inner Sphere are more mature?

My arse, that's why they slowed PPCs AGAIN.

[MadTulip]

Iron Riding Cowboy, on 04 September 2014 - 03:50 AM, said:

Media:Perfect imbalance

Yea to pay something which is imbalanced and gets imbalanced in a different way every now and then is also entertaining so you can find out every two weeks what works best hopefully before anyone else does and exploit that until it becomes common knowledge by trial and error which is when its about time to imbalance it differently again. I also like that. I kinda dislike it if that gets connected to $$$ although even that is to some extend understandable. In that sense the model i triied to give should not be used to balance the game but to find out about the lastest thing to exploit. Maybe i should therefor keep my secret on how to catch the biggest fish although im not good at aiming for myself then .

Anyway guys. lets not get to philosophical here. I made this because i like fitting and plotting things out of pure interest. Raging about the balance is also part of what this is about i guess. The game seems to be fun enough for me to play it for 2 years now and spending time on analysing it instead of shooting.

Edited by MadTulip, 04 September 2014 - 04:07 AM.