mr7q
Destrier
Posts: 71
Karma: 48
Pilot name: mr7q
Platform: Android
|
Post by mr7q on Jan 20, 2017 11:18:03 GMT -5
Going to try to keep this as brief as I can to avoid too much tl;dr-ing, but that may be inevitable. Warning, there are Greek symbols ahead. Disclaimer: I'm only a few months into this game on Android. I write code for a living, and spend a fair bit of my free time gaming (tabletop and recently online) when I'm not outside. I'm no expert in what I'm writing about here, but I try to avoid jumping to conclusions without sufficient evidence, and be constructive wherever possible. I encourage folks to say I'm wrong, as I make no claims on correctness, and will always appreciate critical thought. I also encourage others to think about how they would address the issue of trying to balance experience vs. equipment. As I'm a firm believer that every complaint should be immediately followed by a solution, or you're wasting everybody's time, the following will be my thoughts on how I would do it. As such... Words. The Elo system works as a differential system, using your rank compared to the rank of your competitors and applying the result of your match to your future ranking. Wikipedia does a good explanation if you want to know more. However, three things I think are important: - Starting out, the system is only as good as its initial seed. Your ranking is only ever adjusted when you play.
- Accurate ranking is required for the system to be stable. Because ratings are only adjusted by comparing existing values, errors will take a while to sort out in an established system.
- The ranking is based on the assumption that both players are equally matched in resources. It's an excellent system for Chess, as my pawn is just as capable as Kasparov's.
From this, I am getting the impression that Pixonic is primarily using an Elo style ranking for iOS matchmaking, and I believe this to be a mistake. I am going to make the assumption that wins, and possible damage, are the primary weighting factors for determining player rating. This, I believe, is compounded by the fact that Pixonic improperly weighted the seed values for these player ratings such that it's going to take a long time to stabilize into anything useful. Again, in my opinion only, what has happened is that a ranking system is effectively noise due to the disparity in hangar constructions among varying players such that the system will not be stable for a long time for no meaningful data can be taken from a win/loss result, given that the original ratings that seeded the game can't be trusted to begin with. Consequently, using a win-rate based Elo ranking as your primary matchmaking system is going to end up with exactly what seems to have happened since it does not account for hangar strength. Players are going to be out of their element, whether they're Top Tier beating on L1 Cossacks, or former 4/8 Silver Tier folks trying to face a 12/12 Trident Fury. The original issue that this was trying to solve was that players could min/max the existing tier system to have favorable matchups at all times, usually at the expense of those at the bottoms of those hard tiers. Additionally, using only the highest level bot and its average weapon level further means the system is easily gamed in favor of those willing to do so. As such, I agree with the original assessment that the tier system was ripe for abuse. However, that does not mean that the hangar strength should be either completely, or mostly, ignored in the matchmaking component of the game. I think it goes without saying that there should be a hangar strength metric which is used as the primary matching system, with Elo being used to adjust that score up and down. I don't know the layout of the player base of War Robots with respect to previously existing tiers as I don't have access to that data, so this is all uner the assumption that the matchmaker will not be starved for matches. Instead of the tier system, I would propose that a different metric be used as the primary matching component. Crudely using math, here's the crux of what I'd propose:  This is intended to be a sum of of a player's hangar, multiplying the health of each bot by the sum total of their weapons' effectiveness over range. The GlobalRangeMultiplier is intended to be a tuning factor that allows the "nominal" effectiveness to be determined such that damage output and range are inversely related. Additional tuning factors would also be added to compensate for shields, special abilities, etc..., but there's already enough Greek letters up there for one time. The Elo score would then be a secondary modifier on the above hangar strength, shifting the rating up or down to increase or relax the challenge an individual sees based on their win/loss rate. It addresses, at least in my mind, the key issue that levels for both bots and weapons do not necessarily translate into damage or the ability to take it. This is further compounded that the previous system doesn't account for number of hangar slots, which is crucially important as one plays the game more. Players would matched with either hangars with enemy HP equal to their own damage output, or enemy DPS comparable to their own HP pool. I have yet to do some primary calculations regarding edge cases (i.e. a top tier player using a five slot hangar with a 12/12 Trident Fury and a four 1/1 Destriers) since that would be a lot of data input I'm not willing to do (unless someone has all the level/HP/DPS data already collated into csv somewhere). Oddly enough, the above math isn't too far from what one of Pixonic's developers has already written about here: medium.com/@hex3r_/the-damage-is-too-damn-high-or-achieving-the-perfect-balance-3ccccbe70756#.ypx15tahtIn conclusion, I'm eager to see how the iOS Matchmaker is adjusted with respect to what I believe to be a broken system, and how that eventually translates over to the Android implementation. Top tier players gloating over Clubbers comeuppance aside, the feedback for the current implementation has been overwhelmingly negative, and this kind of bad hype can be extremely detrimental to a market where another game is only a click away. As such, I have my fingers crossed that Pixonic will quickly address the issue. Until then, however, I'm going to feel like I'm on borrowed time, enjoying challenging but fair matches in High Silver on the Android while they last. |
|
|
|
Post by zman on Jan 20, 2017 12:27:47 GMT -5
Honestly I think we could run a much better system. We really don't need a massive global ranking system, or even Elo. I think a Level>WinRate>HangerStrength>Bot Strength system could work very well. Simple yet elegant.
Match Making Strength= 2xLevel + HangerStrength +(HangerStrength x WinRate)
Match Making Strength is used to create a balanced match. When someone initiates a game, they are either added to the pool of currently searching players and the MM groups players by like MMStrength into roughly equal teams. This system is designed with Pressure towards a 50% winrate, but good pilots will often exceed their Hanger Strength and routinely do well, this is reflected in their Win Rate which adds a small penalty the further above 50% they are. Does this mean everyone will have a 50% win rate, no, because good players will play above their hanger strength and the system will find a balance point of difficult where they will maintain their win rate. Good players will maintain win rates above 50 or 60 or even 70%, and poor players will be below 50%. A player's winrate will thus become reflective of the pilot's skill. Some safeguard against players tanking their winrates by not playing would need to be implemented. Using the last 100 or 200 previous matches would be better than the previous 50 matches for this.
Bot Strength
Use basic calculation to calculate Bot Strength.
Bot Strength = Base Strength x Class Coefficient
Base Strength = 10x(Avg(Bot Level +AvgWeaponLevel+AvgWeaponLevel))
Class Coefficients Light: 1.0; Med: 1.1; Heavy 1.2.
So a Level 5 Bot with Level 5 Weapons is a Base Strength 50. A Level 3 Bot with Level 9 Weapons is a Base Strength 70. A Bot 11 Weapons 5 is a base strength 70 as well. A 7/7 is a 70 as well. Here are a couple examples of roughly equivalent bots Light(12/12), Med(11/11), Heavy(10/10).
Base Strength varies from 10-120, Bot Strength varies from 10-144.
Hanger Strength
Hanger Strength = 1.0x1stBot + .8x2ndBot +.6x3rdBot +.4x4thBot +.2x5thBot.
Bots are ranked by strength, 1st=Highest 5th=Lowest
Now, the total strength of a Hanger is not determined by the average bot level, but skews towards the most powerful bot or two. Also, hangers of differing sizes also matter, it is difficult to compare a Hanger of 3 to a hanger of 5.
Hanger Strength varies from 10-432.
The total Matchmaking Strength varies from...
3: Lvl 1 Pilot yet to win
to
924: Lvl 30 Pilot with 100% WinRate and Max Hanger of 12/12 Heavy Bots
Does this fix Clubbing? Lets look at High Bronze
High Bronze Player
238 = 40+132+132(.5)=Lvl 20 Pilot with 5/5 Mediums and 3 Slots and a 50% WinRate.
Clubber
443=60+219+219(.75) =Lvl 30 Pilot with 4/9 Lights, 5 Slots and a 75% WinRate
So, according to my new MM the Clubber has almost Twice the Matchmaking Strength of the High Bronze player and these two Pilots should never ever see each other in battle. What would be an appropriate battle for our skilled "Clubber"? Good Question...
438=60+216+216(.75) = Lvl 30 Pilot with 6/6 Heavies, 5 Slots, and a 75% WinRate
451=60+252+252(.55) = Lvl 30 Pilot with 7/7 Heavies, 5 Slots, and a 55% WinRate
436=60+267+267(.4)= Lvl 30 Pilot with 8/8 Heavies, 4 Slots, and a 40% WinRate.
I know that is a lot of information, but that is where I would start in building a Match Maker.
|
|
|
|
Post by ivx on Jan 20, 2017 12:39:38 GMT -5
Or, you know, they could have just weighted bot and weapon levels evenly and pushed the Gepard into Silver where it belongs with Gary, Stalker and Jesse.
...But I guess I'm wrong, since that would mean naively assuming Pixonic spent months of work developing a new MM system just to save baby seals. Which of course wasn't the case.
|
|
[AurN] perfectlyGoodInk
Aurora Clan Moderator
Posts: 729
Karma: 556
Platform: Android
Clan: leader of Aurora Nova [AurN]
League: Gold
Server Region: North America
Favorite robot: Bishop from Aliens, although WALL-E is a close second
|
Post by [AurN] perfectlyGoodInk on Jan 20, 2017 12:39:45 GMT -5
If you use hangar strength primarily, it needs to weight bots accurately. If there are gross inaccuracies in certain bots or weapons, this creates the kind of exploits that are very well-known to users of this forum.
But aside from that, I would tend to agree. From all accounts, the new MM is supposed to use both, but nobody has revealed what the weighting is.
|
|
|
|
Post by zman on Jan 20, 2017 12:52:29 GMT -5
Or, you know, they could have just weighted bot and weapon levels evenly and pushed the Gepard into Silver where it belongs with Gary, Stalker and Jesse. ...But I guess I'm wrong, since that would mean naively assuming Pixonic spent months of work developing a new MM system just to save baby seals. Which of course wasn't the case. It isn't just about saving seals, it is about improving competition and matchmaking across the board. The hard shelves ie Tier caps programmed into the old MM were a major problem. Bot and Weapons should not be equally weighted, it is one of the reasons that led to low level Geps stocking up on high level weapons to be problematic, or Boas, or Shutzes etc. Weapons were constantly being leveled above Bot Levels and that led to certain advantages in the MM which were readily exploitable. If you use hangar strength primarily, it needs to weight bots accurately. If there are gross inaccuracies in certain bots or weapons, this creates the kind of exploits that are very well-known to users of this forum. But aside from that, I would tend to agree. From all accounts, the new MM is supposed to use both, but nobody has revealed what the weighting is. Yep, but overall Pix has done a pretty good job of balancing bots and weapons, 2.5 seems like it'll be another good step in that direction. Competitive players will always use the best that is available to them, but over Pix isn't tossing massively imbalanced stuff at us. |
|
mr7q
Destrier
Posts: 71
Karma: 48
Pilot name: mr7q
Platform: Android
|
Post by mr7q on Jan 20, 2017 12:52:38 GMT -5
zman I like the thinking. My only worry on the base calculation only using weapon level makes it theoretically easier to find the min/max point. A L5 Punisher vs. a L5 Magnum are two separate beasts, as has been demonstrated in the previous system. Granted, a lot of the level balancing should be determined in the introduction of the weapon, but with the difference in WSP vs. Au vs. Ag weapons, such a single number may not be equal between the currencies (which is why I imagine there were penalties in the old system for WSP bots, Britbots, etc...). ivx The issue with hard tiers is that there is always going to be a tipping point before getting bumped into the next tier. That's why you saw Magnum Gepards everywhere. They were the top of the tier, and everybody knew it. You move those the Gepard up a tier based on that reputation, the clubbers will just end up using Thunder Schuetzes (which are even more durable thanks to the recent HP readjustment).
|
|
|
|
Post by zman on Jan 20, 2017 12:59:07 GMT -5
zman I like the thinking. My only worry on the base calculation only using weapon level makes it theoretically easier to find the min/max point. A L5 Punisher vs. a L5 Magnum are two separate beasts, as has been demonstrated in the previous system. Granted, a lot of the level balancing should be determined in the introduction of the weapon, but with the difference in WSP vs. Au vs. Ag weapons, such a single number may not be equal between the currencies (which is why I imagine there were penalties in the old system for WSP bots, Britbots, etc...). But overall, the balance is pretty good. If we can't count a lvl 5 weapon as a level 5 weapon we have problems. Pix has been improving things and old penalties were bandainds for a broken system. With my suggestion they wouldn't be necessary. The only potential problem would be sneaking in lvl 1 weapon onto a multiweapon to offset a single lvl 12. Like a Griffin wielding 2xlvl12 and 2xlvl1 weapons to average out to a 6.5 average. or a Leo with 3xlvl1 and lvl12 to average Weapons out to a 3.75. This is about the only real exploit to hanger Strength I can see, applicable to really only four weapon and some 3 weapons bots, and it really wouldn't be anything compared to what we saw in the old MM. A more complicated fix would be weighting weapons by slot type to get a different average number that weighted a Heavy slot more than a light slot etc. |
|
[AurN] perfectlyGoodInk
Aurora Clan Moderator
Posts: 729
Karma: 556
Platform: Android
Clan: leader of Aurora Nova [AurN]
League: Gold
Server Region: North America
Favorite robot: Bishop from Aliens, although WALL-E is a close second
|
Post by [AurN] perfectlyGoodInk on Jan 20, 2017 13:10:27 GMT -5
I agree on tiers. I used to shoot 9-ball in the APA league which used a tier system with 9 tiers, and all the best-performing players were the ones who knew how to do just "badly" enough (i.e., "miss" a non-defensive shot convincingly) to stay at the top of each tier.
Micro-tiers is somewhat better, but a formula that generates a score with a smooth and continuous graph would be best.
|
|
|
|
Post by zman on Jan 20, 2017 13:22:13 GMT -5
I agree on tiers. I used to shoot 9-ball in the APA league which used a tier system with 9 tiers, and all the best-performing players were the ones who knew how to do just "badly" enough (i.e., "miss" a non-defensive shot convincingly) to stay at the top of each tier. Micro-tiers is somewhat better, but a formula that generates a score with a smooth and continuous graph would be best. I completely agree, which is why my suggestion creates a continuous curve with enough granularity in it. Best part is no matter what hanger strength you field you'll be facing difficult enough matches that every match should be a battle to maintain your win percentage, and that win percentage is earned and maintained with skill not just hanger. It wouldn't be rubberbandy, and should keep everyone contributing. It also by default keeps the newbies in their pool long enough to grow out of it without ever feeling like they crossed a hard threshold, like the I just bought my first Heavy and moved From bronze to silver feeling you got under the last MM. |
|
|
|
Post by ivx on Jan 20, 2017 13:30:32 GMT -5
The current system honestly isn't any better than the old one; in many cases and instances it's a lot worse for even more players than the previous one (myself included).
We've all discussed this to no end and I don't want to re-write everything but basically anyone caught below a certain level suffers for really no good reason. Even if they don't necessarily lose, they don't end up contributing anything to the game because their team is way above them. I've seen some estimates of 30-50 games to find your new ranking and I've played about a 100 or so games since the MM dropped - I am still grossly mismatched in most (not a figure of speech - truly MOST by far, like 8/10) of my matches. Outclassed by opponents far, far, far above me most of the time. Now I'm even taking detailed post-match screenshots for the last twenty or so matches to prove it if needed later. And many others are in the same boat. Then there are issues with bot types, half the AU purchases suddenly being worthless long-term (where before they could be used to play in a certain tier and the game had nice diversity of a more static type) etc.
Would dropping soft or hard tiers on top of the current new MM solve this to some degree?
Is it theoretically possible to keep the current matchmaking system and drop tiers so that people play with Elo but within their tier? So that the best in their weight category aren't simply crushed by unfair matches with the next category as a "reward" for being the best, and the rest get to play against similarly (un)skilled people until they learn to be better?
|
|
[AurN] perfectlyGoodInk
Aurora Clan Moderator
Posts: 729
Karma: 556
Platform: Android
Clan: leader of Aurora Nova [AurN]
League: Gold
Server Region: North America
Favorite robot: Bishop from Aliens, although WALL-E is a close second
|
Post by [AurN] perfectlyGoodInk on Jan 20, 2017 13:47:30 GMT -5
I really don't think an MM should take too many matches to figure it out. If it's really based on the Bayesian TrueSkill system as has been suggested elsewhere here, the MM should update its prior probabilistic assumptions with every match. A large mismatch between the results and MM’s expectations should result in a large adjustment in its future expectations, especially if its prior assumptions were not based on very much data. Likewise, once a player has established a stable skill level, a single match or even a few matches that do not match the MMs expectations would be weighed against the large history of data and thus have a relatively small effect, making it difficult to game the system. This would make the optimal approach of a player to aim for steady improvement, as this will allow MM to build a stable estimate. Its slow adjustment afterwards should allow them to beat its expectations regularly. So, I think the Devs must be doing something wrong. |
|
|
|
Post by zman on Jan 20, 2017 13:51:47 GMT -5
The current system honestly isn't any better than the old one; in many cases and instances it's a lot worse for even more players than the previous one (myself included). We've all discussed this to no end and I don't want to re-write everything but basically anyone caught below a certain level suffers for really no good reason. Even if they don't necessarily lose, they don't end up contributing anything to the game because their team is way above them. I've seen some estimates of 30-50 games to find your new ranking and I've played about a 100 or so games since the MM dropped - I am still grossly mismatched in most (not a figure of speech - truly MOST by far, like 8/10) of my matches. Outclassed by opponents far, far, far above me most of the time. Now I'm even taking detailed post-match screenshots for the last twenty or so matches to prove it if needed later. And many others are in the same boat. Then there are issues with bot types, half the AU purchases suddenly being worthless long-term (where before they could be used to play in a certain tier and the game had nice diversity of a more static type) etc. Would dropping soft or hard tiers on top of the current new MM solve this to some degree? Is it theoretically possible to keep the current matchmaking system and drop tiers so that people play with Elo but within their tier? So that the best in their weight category aren't simply crushed by unfair matches with the next category as a "reward" for being the best, and the rest get to play against similarly (un)skilled people until they learn to be better? Seriously dude, how many times do I have to explain to you that the problem isn't you playing enough games to find your Elo, it is waiting for everyone else to play enough games to find theirs. The MM is constantly sifting people people and this process takes time. Those running lower strength hangers with lots of skill are the ones who are going to feel the brunt of this, and it sucks, but will very likely only be temporary. We are only like 48hrs in, I know things have settled down quite a bit up by me, but there are still a couple of people in each mach that are too strong or too weak, but overall the games themselves are pretty close, and that is a good sign. One problem you might end up having is that there are not enough people with your skill playing your low powered hangers and you will get shuffled up until the system puts you where it thinks you'll have competitive matches. Are you placing in the top three on the winning side in a quarter of your matches? If so that is "balanced" and skill doesn't matter, but if you are holding a winning percentage and are in the top three more often than not then skill still matters and it is a question of degree. Soft or hard tier caps are what leads to exploitation and if are implemented need to be transparent. Problem is people will do whatever they can to give themselves a potentially unfair advantage and optimize to the top of the tier. We've heard rumors of Leagues that may do just this. Without actual "leagues" or "tiers" you adhere to certain rules to play in a granular system is best. Rembmer, for every player out there stomping people and winning 80+% of their matches and being number one on their team there are multiple people who feel like they never have a chance. This MM may lean too far towards balanced matches, but we don't have enough data yet. Keep playing, and I bet it gets better each day. I bet today isn't as bad as Wednesday was? |
|
|
|
Post by Trogon on Jan 20, 2017 13:58:16 GMT -5
Even if things do eventually even out for my "main" hangar, the one thing that is glaringly apparent is that changing up hangars to play in different tiers (not to club, for variety of play) does not work with this new MM. I face exactly the same types of matches when I switch to my low-level Ag lights as I do with my mid-level heavies. I assume if I played long enough, it would balance out again for my new hangar, but that's not what I'm looking for. It's a big loss to not be able to jump tiers at will for those of us who enjoyed that aspect of this game.
|
|
|
|
Post by zman on Jan 20, 2017 13:58:38 GMT -5
I really don't think an MM should take too many matches to figure it out. If it's really based on the Bayesian TrueSkill system as has been suggested elsewhere here, the MM should update its prior probabilistic assumptions with every match. A large mismatch between the results and MM’s expectations should result in a large adjustment in its future expectations, especially if its prior assumptions were not based on very much data. Likewise, once a player has established a stable skill level, a single match or even a few matches that do not match the MMs expectations would be weighed against the large history of data and thus have a relatively small effect, making it difficult to game the system. This would make the optimal approach of a player to aim for steady improvement, as this will allow MM to build a stable estimate. Its slow adjustment afterwards should allow them to beat its expectations regularly. So, I think the Devs must be doing something wrong. But, there can be a lot of noise in the data in a team match to figure out how well each performed and update their data appropriately. A mediocre player that gets paired with an amazing player on the winning team gets an upward bump, but doesn't deserve one, while a great player on a poor team that loses gets dropped down and moves the wrong direction. Throw in people just ejecting from games and trying to rate all the new people dropping into the system with hanger strengths all over the board and it is pretty wild. I mean, one or two strong guys carry a team beating an appropriate team and the bad guys get bumped up and the balanced guys all get dropped. This is what takes so many games as the MM is turbulent and it can sometimes take a lot of data in a team game to rate each player on that team. An example I saw was having two guys eject and not play in two back to back matches. Despite being against appropriate I then took two losses, the system would lower my Elo because of that when it was undeserved. |
|
|
|
Post by zman on Jan 20, 2017 14:00:15 GMT -5
Even if things do eventually even out for my "main" hangar, the one thing that is glaringly apparent is that changing up hangars to play in different tiers (not to club, for variety of play) does not work with this new MM. I face exactly the same types of matches when I switch to my low-level Ag lights as I do with my mid-level heavies. I assume if I played long enough, it would balance out again for my new hangar, but that's not what I'm looking for. It's a big loss to not be able to jump tiers at will for those of us who enjoyed that aspect of this game. Not necessarily, your low hanger with your established Elo could very well be throwing you up against people without established Elos and stronger hangers. It should adapt to hanger strength eventually, but that may been time to work itself out. |
|
[AurN] perfectlyGoodInk
Aurora Clan Moderator
Posts: 729
Karma: 556
Platform: Android
Clan: leader of Aurora Nova [AurN]
League: Gold
Server Region: North America
Favorite robot: Bishop from Aliens, although WALL-E is a close second
|
Post by [AurN] perfectlyGoodInk on Jan 20, 2017 14:06:33 GMT -5
I really don't think an MM should take too many matches to figure it out. If it's really based on the Bayesian TrueSkill system as has been suggested elsewhere here, the MM should update its prior probabilistic assumptions with every match. A large mismatch between the results and MM’s expectations should result in a large adjustment in its future expectations, especially if its prior assumptions were not based on very much data. Likewise, once a player has established a stable skill level, a single match or even a few matches that do not match the MMs expectations would be weighed against the large history of data and thus have a relatively small effect, making it difficult to game the system. This would make the optimal approach of a player to aim for steady improvement, as this will allow MM to build a stable estimate. Its slow adjustment afterwards should allow them to beat its expectations regularly. So, I think the Devs must be doing something wrong. But, there can be a lot of noise in the data in a team match to figure out how well each performed and update their data appropriately. A mediocre player that gets paired with an amazing player on the winning team gets an upward bump, but doesn't deserve one, while a great player on a poor team that loses gets dropped down and moves the wrong direction. Yes, that there is variation and outlier matches is exactly why a wealth of historic match data should be weighted more heavily than the latest match results according to a Bayesian perspective (which also holds that the frequentist approach commonly taught in Intro Stats overweights the latest results). |
|
|
|
Post by Trogon on Jan 20, 2017 14:09:28 GMT -5
Even if things do eventually even out for my "main" hangar, the one thing that is glaringly apparent is that changing up hangars to play in different tiers (not to club, for variety of play) does not work with this new MM. I face exactly the same types of matches when I switch to my low-level Ag lights as I do with my mid-level heavies. I assume if I played long enough, it would balance out again for my new hangar, but that's not what I'm looking for. It's a big loss to not be able to jump tiers at will for those of us who enjoyed that aspect of this game. Not necessarily, your low hanger with your established Elo could very well be throwing you up against people without established Elos and stronger hangers. It should adapt to hanger strength eventually, but that may been time to work itself out. I hope you're right. |
|
|
|
Post by ivx on Jan 20, 2017 14:52:19 GMT -5
Seriously dude, how many times do I have to explain to you that the problem isn't you playing enough games to find your Elo, it is waiting for everyone else to play enough games to find theirs. Whoops, my mistake - no need to explain anything, was already perfectly understood and I thought I mentioned it. I've posted taking this into account in another thread, somewhere here in General I think. What I said was that I've played about a 100 now and a person posting above me said they played 200. And that if it takes many matches for each and every player in the database, that it may take weeks or months for it all to settle. A relatively long time as mobile games go. Are you placing in the top three on the winning side in a quarter of your matches? If so that is "balanced" and skill doesn't matter, but if you are holding a winning percentage and are in the top three more often than not then skill still matters and it is a question of degree. I'm currently hovering around 66-70%, but rarely in top three - mostly carried by my team who are way above. In some matches I am top, mostly in cases where people either obviously don't play well (Rhinos walking around turning their 「multiple dookie delivery chute」 in Shenzhen central plaza) or we get matched with low levels. So this "good" War Robots isn't really a reflection of my skill as much as of misplacement into matches between higher level players where the best I can generally do is sacrifice bots for beacons. In other words, no amount of skill gets me anywhere if so much as peeking around a corner to check the beacon gets me one-two shotted by a Trebuchet, Trident, PDB, RDB or Zeus levelled way above anything I use. Keep playing, and I bet it gets better each day. I bet today isn't as bad as Wednesday was? I'm sorry but not so far. Nope. Still getting considerably more powerful opponents in most matches - I'd say 70% of matches if we count 11s and 12s (vs. my 4 bot and 6 or barely a few 8 weapons), 80-90% if we count 9s and 10s too. I mean both bots and weapons. Maybe it's just bad luck and playing with people who haven't played that much yet. But it's a huge majority of matches. What I've noticed though is that team powers do seem to be balanced, but in two general categories. You have the top two or three which are generally matched with each other well and then the bottom three-four which are again matched with each other. But the problem is that there's a x 2 bot/level difference between them in the same match. |
|
|
|
Post by kchan4487 on Jan 20, 2017 22:46:17 GMT -5
With the question of how I would do it, I would base it on a combination average damage and average hp of all bots in the hangar.
We can base the 'skill' of the player on average damage dealt since a skilled player with lower level bots and weapons will have higher than average damage than those with similar strength bots, while the less skilled players are likely to deal lower damage compared to the average of those with the same bots. So we're basically pitting people with similar potential dealt damage to each other.
To keep this 'fair', we'd also want to have their entire hangar have close enough average HP across bots. This would allow you to run an all tank hangar, or all light, or a nice mix.
This would let people fight opponents with similar damage potential, yet similar overall hangar durabilities. This would also eliminate seal clubbing because the disparity in damage dealt and average hp would keep them off the less equipped population.
This would also keep a semblance of a tier system where people can stay within a certain tier where there is certain average damage and hp of bots. It would make it hard to drop tiers though since your tier is governed by the average damage over a period of games so you'll have to tank.
Then we decrease the average taking from 50, to 5 games, so that if anyone tanks, he quickly gets thrown up again to higher tiers.
It's simple, easy to maintain current tier, easy to climb up, difficult/troublesome to climb down and club.
Of course there may be better ideas than this.
|
|
|
|
Post by Dredd77 on Jan 21, 2017 0:17:26 GMT -5
Might have missed it, but haven't seen making bot level a cap to weapon level proposed yet. With that loophole closed, restore tier play with a mm system that matches approximate opposition within that tier.
|
|
|
|
Post by boomsplat on Jan 21, 2017 0:34:26 GMT -5
Even if things do eventually even out for my "main" hangar, the one thing that is glaringly apparent is that changing up hangars to play in different tiers (not to club, for variety of play) does not work with this new MM. I face exactly the same types of matches when I switch to my low-level Ag lights as I do with my mid-level heavies. I assume if I played long enough, it would balance out again for my new hangar, but that's not what I'm looking for. It's a big loss to not be able to jump tiers at will for those of us who enjoyed that aspect of this game. IMO, it would've taken some of the sting out of this new MM if Pixo would have also deployed the ability for custom matches. Want to run your 4/9 Cossack army against other like minded individuals? Great - find the players and do the match. Same with those who are growing tired of the MM craziness....it would have been closer to league type play but with willing opponents. at least I think it would've kept more players engaged as we work thru the MM. wishes and dreams.... |
|
Deleted
Deleted Member
Posts: 0
Karma:
|
Post by Deleted on Jan 21, 2017 1:58:01 GMT -5
MM: Hangar + Elo
The traditional Elo system, once up and running, usually scores players between 500 and 2500, with anything breaking either side being vicariously terrible or amazing respectively(28XX being the highest rated Chess score iirc). That being said, the former Hangar rating system was based on a 75-1015, if combined with Elo, creates a sliding metric of (roughly) 600 to 3500.
Hangar score: Decided by the highest factor between bot and weapon levels (1/6, 6/1, and 6/6 all being the same rating per class)
Lights: 100 base rating, +60 per level, for 760 at lv 12.
Mediums: 240 base rating, +60 per level, 900 rating at 12.
Heavies: 380 base rating, +60 per level, 1040 rating at 12.
This more or less, emulates the previous hanar scoring system.
Elo portion: the elo floor rating for all players is set to 100, so the minimal MM score a player will ever have is their hangar score+100
The hangar scoring part would treat players the same as the previous MM system did, causing little culture shock outside of the New MM treating 4/12 bots as simply lv 12.
Elo generated by the new system would be established and calculated using a modified Performance rating and FIDE elo methodology as follows:
The scores of created teams are respectively averaged to create expected performances for red and blue(outside of a player's matches vs NPCs when they first start, unless every player in the same game is running the same gear, level wise, there is no way to zero sum).
If red has a rating of Rr and B has a rating of Rb, the expected outcome is:
Er= 1÷(1+10^((Rb-Rr)÷400)
Eb= 1÷(1+10^((Rr-Rb)÷400)
The system uses these projected outcomes to rate the teams after the match is over, and assigns points to teams according to(E being expected outcome,A being the actual outcome (0 for loss, 1 for win):
K(A-E)
With K being:
Score:K
>2300: 200(for first week after launch)
>2400: 100
2400+: 50
The distribution from player to player would be a % of the total points gained/lost by the team, based on performance in game. The sole caveat being that a player that is not top 3 that wins the medal of capture recieves the same percentage as 3rd place.
Win:
1st:20%
2nd:18%
3rd:17%
4th:16%
5th:15%
6th:14%
Lose:
1st:14%
2nd:15%
3rd:16%
4th:17%
5th:18%
6th:20%
Example:
Team 1:
Player:Score(Hangar+ 100 floor rating)
P1:200
P2:200
P3:200
P4:340
P5:340
P6:480
Total:1760
Average:294
Team 2:
P1:200
P2:340
P3:340
P4:340
P5:340
P6:340
Total: 1900
Average: 317
E1= .47
E2= .63
Outcomes:
Team 1 wins:
Team 1 recieves
200(1-.47)=
200(.63)= 126 points awarded to team, players individually recieve:
1:25
2:23
3:21
4:20
5:19
6:18
Team 2 recieves
200(0-.63)
200(-.63)= -126 points taken from team, players individually lose:
1:18
2:19
3:20
4:21
5:23
6:25
Team 2 wins:
Team 2 recieves:
200(1-63)
200(.47)= 94 total points
Individual players gain:
1:19
2:17
3:16
4:15
5:14
6:13
And team 1 loses the equal amount in the opposite distribution.
They could also raise the floor to any base number that they deem fair for the K factor to be effective, I used 100, but 1000-1300 would work just as well for the K factor thresholds I used.
Personally, I'd set floors on certain factors such as:
100: New player
500:level 30 player
+200 or each Au slot
New player+ 4 total slots= 500 floor
Lv 30 player w 5 slots = 1100 floor
But all that is just me...
Implementing something like that would have very little impact on low level gameplay, create a diverse middle ground, and have top pilots duking it out if Leagues were implemented through scoring brackets to distribute rewards...
Once again, just my idea of fair..
|
|
Deleted
Deleted Member
Posts: 0
Karma:
|
Post by Deleted on Jan 21, 2017 6:08:44 GMT -5
Bot level,
Bot Type, 5 for light, 10 for mid, 20 for heavy, wsp + 10% gold+20%
Specials - event only ones, add an extra 10% or so.
Weapon Level as above. For a base figure.
Hangar slots, 3 - 0%, 4- 25%, 5- 50% increase on overall points.
Separate the hangar slots into tiers.
Should arrange matches with similar bots, making it more about skill than, "What the f... just killed me!?!? I only walked around the corner."
However, if all snipers, things could get boring mightily fast.
|
|
JstKnwImThad
Destrier
My strawberry yoghurt fills the void in my soul
Posts: 57
Karma: 32
Pilot name: ClMeThad
Platform: iOS
Clan: bMf3
League: Gold
Server Region: Asia
Favorite robot: Galahad
|
Post by JstKnwImThad on Jan 21, 2017 8:59:08 GMT -5
I would take into account bot level, weapon level, bot type, weapon hardpoint type... And also not forgetting winrate and average damage. I'm not sure how to implement all this into a fair matchmaking program, and I personally wouldnt like this too much as well. I'm just trying on the developer's shoes.
|
|
mr7q
Destrier
Posts: 71
Karma: 48
Pilot name: mr7q
Platform: Android
|
Post by mr7q on Jan 21, 2017 10:30:03 GMT -5
The more I think about it, and the more I see human nature revealed as folks figure out how to abuse an Elo system, I'm now of the opinion that the basis of the matchmaker should be cumulative hangar strength (multiple implementations discussed above) plus an exponential bonus for the amount your win rate is above 50%. For example (notional values), a 50% win rate would be a 1x multiplier on hangar strength, a 60% win rate a 1.2x, 70% 1.5x, 80% 1.75x, and 90% 2x. I'd need to do more math than I'm willing to do on a Saturday morning to optimize that curve, so don't take that slope as my direct suggestion. This would prevent a player from being able to tank their hangar strength rating by intentionally losing to fool the system, while a more granular hangar strength system would prevent new players from having the shiny new bot they bought immediately booting them up a tier for which they're unprepared.
|
|
whitecrow666
GI. Patton
Posts: 144
Karma: 36
Pilot name: whitecrow666
Platform: Android
Clan: 2017 war Gods
League: Gold
Favorite robot: Gareth
|
Post by whitecrow666 on Jan 21, 2017 10:39:40 GMT -5
More game modes would solve lot issues.
There should be a capture flag mode .
Free for all.
Light bots only mode.
And the game needs more maps.
I have no problem with new match makeing
.
But they i never was a getard
|
|
bearre
Destrier
Posts: 122
Karma: 78
Pilot name: Bearre
Platform: Android
Clan: None
League: Silver
Favorite robot: RDB Griffin
|
Post by bearre on Jan 21, 2017 11:29:12 GMT -5
The game is good thus far; the art work, interface, weapon and bot designs are top notch. The weak point and source of many complaints has been a lack of balance in the mm system.
Originality is great but some systems are used repeatedly because they work well. A common method in pvp games is to limit the weapon level to the base machine. Thus a level 4 bot can only have level 4 and below weapons. This allows the devs to create level based tiers 1-3 4-6 7-9 10-12. When your highest bot reaches lev 4 you advance to the 2nd tier. This elliminates folks with level 4 bots and lev 12 weapons "seal clubbing" lower levels and means at worst when entering a new tier you will only be out gunned by a few levels at most, no facing lev 12/12 bots in 6/7 machines.
As many have weapons over their bot levels perhaps you can code it so that a gun leveled higher than the machine it is on will perform at the level of said machine. A lev 12 mag on a lev 4 bot functions at lev 4. This will allow you to take your weapons off your top tier machines and through them on lower lev bots to squad with clan mates in lower tiers. I'm not sure how hard this would be to code but I am sure you could do it, you design missile systems so you guys really are rocket scientists?
I'm sure some folks will find issues to complain about but I have seen this work well in many similiarly themes games and seems to insure the matches are fairly even and decided by skill luck and teamwork vice grossly uneven forces.
|
|
|
|
Post by lilryry on Jan 22, 2017 18:46:07 GMT -5
The game is good thus far; the art work, interface, weapon and bot designs are top notch. The weak point and source of many complaints has been a lack of balance in the mm system. Originality is great but some systems are used repeatedly because they work well. A common method in pvp games is to limit the weapon level to the base machine. Thus a level 4 bot can only have level 4 and below weapons. This allows the devs to create level based tiers 1-3 4-6 7-9 10-12. When your highest bot reaches lev 4 you advance to the 2nd tier. This elliminates folks with level 4 bots and lev 12 weapons "seal clubbing" lower levels and means at worst when entering a new tier you will only be out gunned by a few levels at most, no facing lev 12/12 bots in 6/7 machines. As many have weapons over their bot levels perhaps you can code it so that a gun leveled higher than the machine it is on will perform at the level of said machine. A lev 12 mag on a lev 4 bot functions at lev 4. This will allow you to take your weapons off your top tier machines and through them on lower lev bots to squad with clan mates in lower tiers. I'm not sure how hard this would be to code but I am sure you could do it, you design missile systems so you guys really are rocket scientists? I'm sure some folks will find issues to complain about but I have seen this work well in many similiarly themes games and seems to insure the matches are fairly even and decided by skill luck and teamwork vice grossly uneven forces. absolutly the best idea I've heard on this subject, solves all the problems with the previous system as well as the abortion that is the current mm system |
|
|
|
Post by Dredd77 on Jan 22, 2017 19:08:05 GMT -5
MODERATION: Moved bearre 's excellent inaugural post to the larger, on-topic post. I really like this idea too, bearre. I've thought about the "weap level <= bot level" approach, but yours addresses the inevitable backlash much better.
|
|
|
|
Post by boomsplat on Jan 22, 2017 19:17:31 GMT -5
This system which I will patent and sell to Pixo is the solution for our problems....behold, the Random Squad Generator. 100% natural and quite delicious when it's ready to be replaced by a new updated system (swap chickens if you aren't having good matches). edit: just found out this is not a new idea - this is the new matchmaker currently in use on IOS. They are on their 12th chicken trying to get it right but they will be running out of chickens named Elo here shortly.  |
|
![[AurN] perfectlyGoodInk Avatar](http://storage.proboards.com/6607038/avatar/40AmBXNiR1xsOVmOvxBp.jpg)
