72 replies to this topic

[Bamboozle Gold]

Non-serious / serious post here.
If I in 2021 were tasked with designing a match maker system that balances two teams of varying skill levels and different mechs I would try if a neural network would be a good fit for the problem. Neural networks are great at taking multiple inputs data and giving a simple solution. That's one of the things they're great at, in fact.

We could give it the last 100 matches of every player (mech played, damage done, damage dealt, kills, assists, win / loss, friendlies, the whole shebang) and ask it to balance the teams so it's likely to produce an even score. Considering this is very little data compared to image or video recognition it would run reasonably fast as well.

No need to mess with PSR or whatever, just let the magic blackbox deal with balancing the teams. Sure, there would likely be some hilarious consequences (like for example maybe it always places that 1 Shadowcat against 3 slow assaults because that gives better numbers) but overall it would probably be decent.

[Rkshz]

MrMadguy, on 21 September 2021 - 10:03 PM, said:

What WLR MM adepts don't understand - is that MM should measure and balance PERSONAL skills.

I disagree with you
if the players have an WLR of 0.9, this means that this player cannot play well
if a player has a WLR of 2.0 this means that the player is playing well, he does everything to win

accordingly, if in one team the players have a total WLR of 2.0 and in the other team the players have a total WLR of 0.9, then which team will win it is obvious

personal skill matters, but in a 12v12 game with equal skill players, this will not have a significant advantage

therefore, the MM balance in terms of WLR is the most objective indicator

Edited by Rkshz, 22 September 2021 - 03:49 AM.

[MrMadguy]

What I talk about - is that WLR MM doesn't even take into account, what actually happens in certain match. Why one player gets 6 kills and other dies early with almost 0 dmg. Because average outcome doesn't show it. This is exact reason, why this game is bad. You're put into one team with pros, who can win without you and you're just stomped. Match after match. And in case of WLR MM I would stuck in this situation for ages.

Edited by MrMadguy, 22 September 2021 - 04:46 AM.

[John Bronco]

You're currently stuck in this situation (well not you personally, since there aren't any 'pros' in tier 4 matches).

[Rkshz]

MrMadguy, on 22 September 2021 - 04:40 AM, said:

What I talk about - is that WLR MM doesn't even take into account, what actually happens in certain match. Why one player gets 6 kills and other dies early with almost 0 dmg. Because average outcome doesn't show it. This is exact reason, why this game is bad. You're put into one team with pros, who can win without you and you're just stomped. Match after match. And in case of WLR MM I would stuck in this situation for ages.

one player kills 6-10 opponents for two reasons:

1. When there are weak opponents in the 1st team, who either stand in one place and do not know what to do, or who do not know how to shoot accurately (such players cannot have WLR higher than 0.9)

2. When in the 2nd team, there are skillful opponents that severely damaged the mechs of the 1st team, and one skilful player simply finishes off these mechs.

In most cases, this is exactly what happens.

In my example: 90% of my "ace of spades" were made on weak opponents (how do I know this? because they cannot dodge and spread damage).

[Nightbird]

MrMadguy, on 21 September 2021 - 10:03 PM, said:

And here is when you make some assumptions. Like "X difference in PSR between teams = K * X chance to win match". And this assumptions are usually about how MM SHOULD WORK - not about how it REALLY WORKS.

For example. Simple question. Do you take into account, that there is constant flow of players in this game? As I can see, you assume, that exactly the same players play this game together for 200k matches. Do you understand, that in reality it doesn't work this way? Or another example: do you understand, that different players can have different time zones, so they just can't play together?

This is where reading before you type can help you. 1. I did not assume the difference in PSR difference is the real chance for a team to win. 2. I used the stats from the player base to simulate my results, and I modified the chance someone will be in queue based on the number of games they play.

Don't fall for the Dunning-Kruger effect.

[dubstep albatross]

Rkshz, on 22 September 2021 - 06:51 AM, said:

one player kills 6-10 opponents for two reasons:

1. When there are weak opponents in the 1st team, who either stand in one place and do not know what to do, or who do not know how to shoot accurately (such players cannot have WLR higher than 0.9)

2. When in the 2nd team, there are skillful opponents that severely damaged the mechs of the 1st team, and one skilful player simply finishes off these mechs.

In most cases, this is exactly what happens.

In my example: 90% of my "ace of spades" were made on weak opponents (how do I know this? because they cannot dodge and spread damage).

In my case, I got my Ace while running a short range brawler Spirit Bear. It was on Crimson and my team went up the backside of the saddle and happened to top it while the other team was at the bottom side, under the roof. A large firefight, with lots of trading, ensued from that position. I opted to just go through the tunnel. When I arrived, I had the enemy team at the side/back and I just plowed through them with SRMs and LBX. I got many backs and, when they finally realized what was happening, they turned to face and were already soft. SRM and LBX are crit and armor-hole seekers.

I could not have gotten my Ace without the help of my team forming a line and I got lucky because I was somewhere totally unexpected, using MASC to make my Spirit Bear move faster than most would guess. I could easily have been picked off because I was isolated and I could easily have exited the tunnel to a team facing me. My build was incredibly short ranged and specific.

All in all, my Ace was due to the right conditions at the right time, made possible by a risky move on my part and my team's competence.

Edited by dubstep albatross, 22 September 2021 - 10:36 AM.

[Rkshz]

dubstep albatross, on 22 September 2021 - 10:12 AM, said:

All in all, my Ace was due to the right conditions at the right time, made possible by a risky move on my part.

for beginners, the first ace is luck - it just happened to be at the right time and in the right place

but when you set yourself the goal of playing well, and you systematically make 6-10 kills, you begin to understand that in 90% of cases it only works against weak teams
a strong enemy team will not allow you to make many kills, this will be rare - for this you need to take risks and be impudent

maybe someone else has it, but after 20 ace I came to this conclusion

[MrMadguy]

I don't know, what is broken in balance in this game, but I really start to feel, that something is broken intentionally. I've been playing this game for several hours today, because I need that 2 last kills/KMDDs for my free 'Mech. Guess what? Zero victories. Zero kills. All matches were 2..3-12 stomps. In Tier 4? It really looks like Tier 1 vs Tier 5 matches, where my team is always Tier 5 and have no clue, how to play this game.

I know, that it's conspiracy theory, but is it because I play unskilled 'Mechs only? I haven't been playing them for a long time, but started recently due to running out of GSP. And it looks like devs have some hidden matchmaking strategies in order to encourage their players to spend MC on converting XP into GXP. Such "balance" just can't be fixed, no matter, what MM would be used.

And again. One can always predict, that some problems will happen in next match. As always, the best indicator of future problems - is voting for "bad" map, when better options are available. It usually clearly shows, that guys: 1) Have no clue, what they're doing 2) Know too well, what they're doing. Anyway. 6x map vote multiplier says everything about current state of this game.

Edited by MrMadguy, 28 September 2021 - 09:53 AM.

[dubstep albatross]

MrMadguy, on 28 September 2021 - 08:29 AM, said:

I don't know, what is broken in balance in this game, but I really start to feel, that something is broken intentionally. I've been playing this game for several hours today, because I need that 2 last kills/KMDDs for my free 'Mech. Guess what? Zero victories. Zero kills. All matches were 2..3-12 stomps. In Tier 4? It really looks like Tier 1 vs Tier 5 matches, where my team is always Tier 5 and have no clue, how to play this game.

Check out the chart below. It's from around 730 games I have played, mostly in tier 4, but some in tier 5 and tier 3. The chart shows the number of surviving pilots on the victory team when the defeated team loses all pilots. This is the most common outcome, by the way. Only about ten percent of my data ends with one or more surviving member on the defeated team.

About 50% of the game data I have shows eight or more surviving members of the victorious team. A 12-3 or 12-4 end game result is quite common. Intuitively, I think there are two main factors for these common results. First, the "snowball" nature of the game. Loss of mechs results in a net loss of armor, targets, and outgoing firepower. Second, imbalances introduced during matchmaking. This may be a flawed matchmaker, variables the matchmaker cannot balance, or both.

Quote

I know, that it's conspiracy theory, but is it because I play unskilled 'Mechs only? I haven't been playing them for a long time, but started recently due to running out of GSP. And it looks like devs have some hidden matchmaking strategies in order to encourage their players to spend MC on converting XP into GXP. Such "balance" just can't be fixed, no matter, what MM would be used.

Unskilled mechs are at a significant disadvantage simply because most mechs on the battlefield are skilled to some degree, if not to max. The baseline on the battlefield is a fully skilled and optimized mech, anything less is just disadvantaged. That creates quite a power difference between unskilled mechs and skilled mechs, especially when the tree is optimized well and harmonized with loadout.

If you want a demonstration of the stark difference, you can try this: do not skill your mech until you have 91 SP available. Play it the entire way up to that much XP, then skill the mech. You will suddenly feel like a god striding upon the battlefield. I think most people incrementally upgrade, either one by one or in short bursts of upgrades. It is much easier to not notice those incremental changes.

An even better demonstration would be to get a duplicate mech, completely unskilled, for your favorite and most effective mech. An alternative could be to re-spec, but not re-distribute the points (I am not sure if that can be done). Play that mech for 10-20 games, completely unskilled.

Edited by dubstep albatross, 28 September 2021 - 10:50 AM.

[Brizna]

Stop talking about particular cases, we are talking about statistics. For every one match a bad pilot kills 7 foes, there's a gazillion matches he kills nothing.

Sure I can lose a match because my team sucks, but if my W/L ratio is 1.3 that means I am just an slightly above average pilot, currently MM thinks I am one of the best pilots in the game. THIS IS A PROBLEM.

So please just STOP bringing single matches, they are just an insignificant grain of sand in STATISTICS.

Edited by Brizna, 28 September 2021 - 11:27 AM.

[MrMadguy]

dubstep albatross, on 28 September 2021 - 10:50 AM, said:

Unskilled mechs are at a significant disadvantage simply because most mechs on the battlefield are skilled to some degree, if not to max. The baseline on the battlefield is a fully skilled and optimized mech, anything less is just disadvantaged. That creates quite a power difference between unskilled mechs and skilled mechs, especially when the tree is optimized well and harmonized with loadout.

If you want a demonstration of the stark difference, you can try this: do not skill your mech until you have 91 SP available. Play it the entire way up to that much XP, then skill the mech. You will suddenly feel like a god striding upon the battlefield. I think most people incrementally upgrade, either one by one or in short bursts of upgrades. It is much easier to not notice those incremental changes.

An even better demonstration would be to get a duplicate mech, completely unskilled, for your favorite and most effective mech. An alternative could be to re-spec, but not re-distribute the points (I am not sure if that can be done). Play that mech for 10-20 games, completely unskilled.

I knew, that somebody would say, that it was my unskilled 'Mech, that was dragging whole team down and caused 2-12 stomp as result. No. Difference between skilled and unskilled 'Mech isn't that big. I have slightly less mobility, slightly less heat efficiency and slightly less QOL. But overall 4xER-PPC Assualt is still good and can still deal 700dmg. On average it's more than enough for Tier 4. It just my teams, that were terribly bad. I had two cadets in last match of that day.

[dubstep albatross]

MrMadguy, on 30 September 2021 - 01:43 AM, said:

I knew, that somebody would say, that it was my unskilled 'Mech, that was dragging whole team down and caused 2-12 stomp as result. No. Difference between skilled and unskilled 'Mech isn't that big. I have slightly less mobility, slightly less heat efficiency and slightly less QOL. But overall 4xER-PPC Assualt is still good and can still deal 700dmg. On average it's more than enough for Tier 4. It just my teams, that were terribly bad. I had two cadets in last match of that day.

Nowhere in my response did I say you were dragging your whole team down and causing a stomp.