30 replies to this topic

[Duke Nedo]

Deathlike, on 16 April 2015 - 11:22 AM, said:

It is literally impossible to use strict mathematical formulas.
...
Mech balance requires actual understanding... and not stupidly strict min-maxing, but actual practical usage in real games.

It depends on your purpose... I am not trying invent some magical formula here that will spit out the meaning of mech balance. But, PGI based their first quirk pass on a Tier list, so the first thing we can try to do is to visualize where we are now after some quirkening. The second thing we can try to do is to do it a bit more accurate than PGI did in the first pass.

Now, the interesting thing we could do with a model that uses input like hardpoint locations etc is to break down why some mechs are considered good or bad. If we also expand that a little, we can actually play with the model to produce answers we are interested in, for example: If we want a quick and dirty list of good brawlers, we can just adjust the weighting for hardpoint location and increase the weighting for durability, sort it and we have a crude list in 20 seconds of work.

I mean, this is what models are... just models. We try to describe reality to some extent using a set of rules, and then use it as a tool.

This is not an ambitious model really, more like a crude test and I'm curious how it would look if I got some good quality input. Perhaps something interesting would pop out, who knows?

[1453 R]

I'd be willing to give it a shot, my weekend's free, but I doubt I'd really qualify for the Experienced Player bracket you're looking at. Certainly not on the level of the Forum Gurus around here. Still, if you're starved for opinions throw me a PM and I can see about working on that this weekend. Be interesting to see what the results are.

[blood4blood]

Armor/survivability.
Firepower.
Speed/maneuverability.

^^^^The three categories that matter for comparison purposes. Assigning numerical values to subjective ratings and trying to use those to objectively balance mechs is just inherently flawed, though. You can make some objective comparisons and assign points on a scale thereby - for example, speed ratings - but there are too many unquantifiable variables to use a numerical rating effectively. (I.e. What number do you assign to Firestarters, Spiders, and Panthers in the armor category to account for how skinny and small they are?)

I think it's an interesting project, but it would have to be done objectively to be truly useful as anything but a point for further subjective discussion.

EDIT: Now if you could get ahold of PGI's databases of mech stats, that could provide some really interesting fodder for mech balance discussions.

Edited by blood4blood, 16 April 2015 - 12:12 PM.

[Deathlike]

Duke Nedo, on 16 April 2015 - 11:51 AM, said:

It depends on your purpose... I am not trying invent some magical formula here that will spit out the meaning of mech balance. But, PGI based their first quirk pass on a Tier list, so the first thing we can try to do is to visualize where we are now after some quirkening. The second thing we can try to do is to do it a bit more accurate than PGI did in the first pass.

Now, the interesting thing we could do with a model that uses input like hardpoint locations etc is to break down why some mechs are considered good or bad. If we also expand that a little, we can actually play with the model to produce answers we are interested in, for example: If we want a quick and dirty list of good brawlers, we can just adjust the weighting for hardpoint location and increase the weighting for durability, sort it and we have a crude list in 20 seconds of work.

I mean, this is what models are... just models. We try to describe reality to some extent using a set of rules, and then use it as a tool.

This is not an ambitious model really, more like a crude test and I'm curious how it would look if I got some good quality input. Perhaps something interesting would pop out, who knows?

There's no shot that I'm interested in creating a pseudo-BV rating for mechs. It's just not happening.

I do have plenty of arbitrary rules.

Head energy hardpoints are "1/2" an energy hardpoint for IS, because you can only fit at best a Medium Pulse Laser on it. Clans have the option of going with a Clan ER Large.. (or Clan ER Med) that not even the IS can do at the current moment.

CT energy hardpoints on an IS mech are treated the same as 1 "full" energy hardpoint (no IS PPC option). 2 Med Lasers (or Med Pulse) is "effectively" similar to a Large/ER Large Laser/Large Pulse Laser.

Both arms totaling 1E (like the Thunderbolt-5S as an example) are generally less useful or "less than effective" for using medium lasers with (Enforcer-5D/Griffin-3M is an exception of sorts due to locality and necessity).

I have other "rules" (particular for missiles and energy), but the point being that it is somewhat arbitrary and on the other hand a way of looking at how things are built based on hardpoints.

Mind you, having a Wolverine-6K is not that much different from a Panther-9R in terms of usefulness and complete reliance on their "gun arm".

So, it is what it is.

[ArchSight]

A scoring system isn't going to work because that's a matter of opinion along with teir lists they are opinion as well. The lists don't prove with number's or facts that make the mech strong. It only supports it by trial and error of what some players have found to work for their success in the role they were doing. Although, some mechs have advantages that are unique causing them to be easily found by a player looking at them.

What needs to happen first before you can start judging which is best is:

To make a list of mechs categorized by all of their stats with a filter that puts it into order of best stat to least best stat upon a person clicking the various stats that they want the best of. Doing so will allow player's to find what mech they want to field for a pacific role they intend to customize a mech for. Every single different Omni-pod combination and the multiple different stats for the number of weapon hard point locations with height is going to be a nightmare to list in this. Referring to the hard point types within that is going to take up some space. The different quirks will all have to have their own stat categories. I don't know how to figure out or put in order the different speeds with all the different engine's for the mechs. Have to get every single different stat into the list.

After the Huge list of mech stats is made, comparisons can start to be made of the builds or different roles that the player's made from the list. Things like: How well it performs on a pacific map, on what game mode it's used for, and how well it performs it's intended role vs other mechs doing the same role. The role or tactical method of applying it's full potential is dependent on what weapons or equipment it has been customized for. Whether the role it's using is viable or not is up to the strategy that the players are performing, not just a tactical discussion. The player or the teams strategy is responsible if the mech load out is not being used the intended way that it's post to function. If player's find a mech that doesn't excel in anything it means PGI has a design fault in their game balance. This will likely cause the mech to be less used than other mechs which is bad if the work that was put into it is being wasted.

Edited by ArchSight, 16 April 2015 - 05:01 PM.

[Duke Nedo]

Deathlike, on 16 April 2015 - 12:46 PM, said:

There's no shot that I'm interested in creating a pseudo-BV rating for mechs. It's just not happening.

I do have plenty of arbitrary rules.

Head energy hardpoints are "1/2" an energy hardpoint for IS, because you can only fit at best a Medium Pulse Laser on it. Clans have the option of going with a Clan ER Large.. (or Clan ER Med) that not even the IS can do at the current moment.

CT energy hardpoints on an IS mech are treated the same as 1 "full" energy hardpoint (no IS PPC option). 2 Med Lasers (or Med Pulse) is "effectively" similar to a Large/ER Large Laser/Large Pulse Laser.

Both arms totaling 1E (like the Thunderbolt-5S as an example) are generally less useful or "less than effective" for using medium lasers with (Enforcer-5D/Griffin-3M is an exception of sorts due to locality and necessity).

I have other "rules" (particular for missiles and energy), but the point being that it is somewhat arbitrary and on the other hand a way of looking at how things are built based on hardpoints.

Mind you, having a Wolverine-6K is not that much different from a Panther-9R in terms of usefulness and complete reliance on their "gun arm".

So, it is what it is.

Great, I respect that you don't want to put any energy into this, that's cool.

But, it somehow seems like some guys here are just dead locked on that some black and white picture of me trying to prove mech tiers by numbers. As I've written many times in these threads, a model is a model and a no model anywhere in the world will be perfect. If it was, it wouldn't be a model.

Again, what I try to do here is to break down the input a bit into a few parameters. They are still subjective, and they have to be as many people pointed out, but by breaking it down a bit it's easier to be systematic about it. My aim is to see how far I can come with this admittedly extremely simple and simplifying model. Could it be useful? I would say surprisingly much so.

Let me illustrate, please try to keep an open mind about what I'm trying to achieve here.

When PGI did the quirkening they based them on subjective Tiers. Let's compare their Tiers with the Tiers that the model spit out when removing the quirk input, i.e. the models view on the unquirked mechs:



Now, read it like this. The model Tiers are in blue, the PGI tiers are in red. The further left in the diagram, the less better the unquirked mech according to the model, and the less quirks it should have received. A large red column means a high PGI tier and strong quirks.

Assaults:
the biggest over estimates are for: STK-5S, STK-4N, STK-3H, BLR-3M, BLR-1G
the biggest under estimates are for: VTR-9S, VTR-9K, VTR-9B, VTR-DS and HGN-733C

Heavies:
the biggest over estimates are for: TDR-9S, TDR-5S, TDR-5SS and CPLT-C4
the biggest under estimates are for: CTF-3D, QKD-5K

EDIT:
Mediums:
the biggest over estimates are for: HBK-4G, HBK-GI, GRF-1S, HBK-4P, BJ-3, BJ-1X
the biggest under estimates are for: none

Lights:
the biggest over estimates are for: SDR-5V, SDR-5K, RVN-2X, RVN-H and interestingly commandos, as well as FS9s.


For these graphs I'd say the the model overestimates the performance of the light lights and perhaps lights in general. Probably because it doesn't really take heat into account in a good way.
//EDIT

Have to run to a lecture, will continue later.

Can't we just agree that using even a simple model like this would have helped PGI make more balanced quirks? And this is not my final point even, the bigger use for a model like this would be to predict the effect of giving various degrees of quirks to different variants... tbc

Edited by Duke Nedo, 17 April 2015 - 12:47 AM.

[Deathlike]

Duke Nedo, on 17 April 2015 - 12:09 AM, said:

Great, I respect that you don't want to put any energy into this, that's cool.

But, it somehow seems like some guys here are just dead locked on that some black and white picture of me trying to prove mech tiers by numbers. As I've written many times in these threads, a model is a model and a no model anywhere in the world will be perfect. If it was, it wouldn't be a model.

Again, what I try to do here is to break down the input a bit into a few parameters. They are still subjective, and they have to be as many people pointed out, but by breaking it down a bit it's easier to be systematic about it. My aim is to see how far I can come with this admittedly extremely simple and simplifying model. Could it be useful? I would say surprisingly much so.

Let me illustrate, please try to keep an open mind about what I'm trying to achieve here.

When PGI did the quirkening they based them on subjective Tiers. Let's compare their Tiers with the Tiers that the model spit out when removing the quirk input, i.e. the models view on the unquirked mechs:



Now, read it like this. The model Tiers are in blue, the PGI tiers are in red. The further left in the diagram, the less better the unquirked mech according to the model, and the less quirks it should have received. A large red column means a high PGI tier and strong quirks.

Assaults:
the biggest over estimates are for: STK-5S, STK-4N, STK-3H, BLR-3M, BLR-1G
the biggest under estimates are for: VTR-9S, VTR-9K, VTR-9B, VTR-DS and HGN-733C

Heavies:
the biggest over estimates are for: TDR-9S, TDR-5S, TDR-5SS and CPLT-C4
the biggest under estimates are for: CTF-3D, QKD-5K

Have to run to a lecture, will continue later.

Can't we just agree that using even a simple model like this would have helped PGI make more balanced quirks? And this is not my final point even, the bigger use for a model like this would be to predict the effect of giving various degrees of quirks to different variants... tbc

Well... conceptually you can do stuff like in baseball to evaluate a player with stats. It's incredible stuff they do in baseball (many of which flies over my head).

Still, I like traditional scouting too... and instincts.

We have to do both to build a better overall picture, instead of a straight numbers game that can be skewed in certain circumstances.

[Duke Nedo]

Deathlike, on 17 April 2015 - 12:13 AM, said:

Well... conceptually you can do stuff like in baseball to evaluate a player with stats. It's incredible stuff they do in baseball (many of which flies over my head).

Still, I like traditional scouting too... and instincts.

We have to do both to build a better overall picture, instead of a straight numbers game that can be skewed in certain circumstances.

I totally agree, this is not an attempt to replace thinking... It's an experiment to see if one could make a tool, a tool to be used critically with caution!

Edited by Duke Nedo, 17 April 2015 - 12:37 AM.

[Duke Nedo]

1453 R, on 16 April 2015 - 12:08 PM, said:

I'd be willing to give it a shot, my weekend's free, but I doubt I'd really qualify for the Experienced Player bracket you're looking at. Certainly not on the level of the Forum Gurus around here. Still, if you're starved for opinions throw me a PM and I can see about working on that this weekend. Be interesting to see what the results are.

Thanks for the offer! Let's wait a bit and see if anyone else is willing to contribute, otherwise it's not fair to have you invest time in it. Somehow I doubt I have the forum-status to pull it off. Too bad, I feel that a data set like that would be interesting by its own right. If one for example reads the topic about top 5 mechs in need of a revamp it's obvious that there is some consensus but also some different opinions. I'd like to see the average with standard deviation... we all have our different kinks I guess.

[Paigan]

You would really have to count in a LOT of things, for example:

- pure weapon builds profit more from modules, are easier to control
- mixed weapon builds automatically go down in efficiency because different weapons are good or bad in different situations.
- ammo weapons have the risk of exploding (both depending on heat generation and via crits)
- TrueDubs vs. ... how are the sucky ones called?
- XL Engine survivability (there's a HUGE difference between Clan XL and IS XL in combat, MUCH more than just 3 points vs. 2 points. You'd have to factor in engine destruction rules, etc.)
- Weapons occupying a lot of slots are much more likely to get destroyed than single slot weapons (whenever I fit an AC20/gauss, is lose them INSTANTLY on the first crit)
- Damage spread (15 concentrated damage of a gauss hurt MUCH more than 20 evenly distributed damage of an LRM-20)
- Also very important: in the end, the viability of a mech depends on the terrain and on team play (e.g. LRM ARE pretty good on open terrain without cover and with decent team play. No skill whatsoever saves you from 1000 incoming missiles if you have nowhere to hide behind)

I'm not saying it can't be done.
I believe TheoryCraft can be viable IF it really properly covers all the variables.

But that's a gigantic IF. Like decades of research (math formulas and stuff) big. You can't just squeeze it into some quick excelsheets and get a nice approximation.

My advice: let it be.

Edited by Paigan, 17 April 2015 - 03:04 AM.

[Duke Nedo]

Paigan, on 17 April 2015 - 03:01 AM, said:

You would really have to count in a LOT of things, for example:

- pure weapon builds profit more from modules, are easier to control
- mixed weapon builds automatically go down in efficiency because different weapons are good or bad in different situations.
- ammo weapons have the risk of exploding (both depending on heat generation and via crits)
- TrueDubs vs. ... how are the sucky ones called?
- XL Engine survivability (there's a HUGE difference between Clan XL and IS XL in combat, MUCH more than just 3 points vs. 2 points. You'd have to factor in engine destruction rules, etc.)
- Weapons occupying a lot of slots are much more likely to get destroyed than single slot weapons (whenever I fit an AC20/gauss, is lose them INSTANTLY on the first crit)
- Damage spread (15 concentrated damage of a gauss hurt MUCH more than 20 evenly distributed damage of an LRM-20)
- Also very important: in the end, the viability of a mech depends on the terrain and on team play (e.g. LRM ARE pretty good on open terrain without cover and with decent team play. No skill whatsoever saves you from 1000 incoming missiles if you have nowhere to hide behind)

I'm not saying it can't be done.
I believe TheoryCraft can be viable IF it really properly covers all the variables.

But that's a gigantic IF. Like decades of research (math formulas and stuff) big. You can't just squeeze it into some quick excelsheets and get a nice approximation.

My advice: let it be.

Heh, I don't get where all this negativism is coming from? Some of you guys just seem to see some graphs and jump to your guns and say that it can't be done, without bothering to read up on what I am actually trying to do.

I am not trying to accurately describe mech performance with a number. I am the first to agree that it cannot be done. Not mainly because of the variables or anything, but because performance is situational. Atlas is a great example, brutal at close range and level ground, useless for long range ridgehumping. So in a general sense an Atlas would be rated in the mid-tiers, simply because there are mechs out there that can do both pretty well, like Stalkers. It would also be rated much higher in a competitive game where your team is more likely to dictate the character of the match etc.

That said, my conclusion from just toying with this is that even a very crude model like this actually does give a surprisingly good correlation with how I perceive mech performance on some sort of general level.

In defense of the model I should also mention that many things that people are criticizing is actually baked into the input. From your points above,

The scores I gave as input are subjective, and take many of these things into consideration. For example, the Hitboxes score can be high for different reasons, XL-friendlyness is one reason (Cent), well distributed CT/ST boxes (Atlas) is another, small CT (Stalker) is a third for STD engine mechs, and slim mech is a fouth (Spider). Hardpoint locality includes both how high they are, and how they are distributed (if you can build asymmetric or if you have to use arms combined with ST). Hardpoint capacity take into account where the HP is (head, CT restricted size or overlapping HP like for the AWS-9M). It's too complex not to use some subjective input here, so the point of this is not to make it 100% by numbers, the point is to break it down a bit into components and to be systematic in rating them.

Truedubs vs Poordubs is not included, that's a good point and I am aware that especially the lighter lights are overestimated. That could be introduced but I am not so sure it's worth the effort right now. There are other equally big flaws that limits how much one should trust the graphs anyways... I mean, the bottom line here is that I have been toying with a tool and as with all tools or models you have to use your judgement when you're interpreting them. Take it for what it is.