Login Register
 
RSS Feed Twitter Facebook YouTube

One step towards somewhat tolerable match making

Home FORUMS Robocraft Discussions General Robocraft Chat One step towards somewhat tolerable match making

This topic contains 14 replies, has 10 voices, and was last updated by  c0l0ssus 3 months ago.

Viewing 15 posts - 1 through 15 (of 15 total)
  • Author
    Posts
  • #2125247

    Alright so I think we’ve all been in these sort of matches multiple times in a row:

    The games looks like this 99.9% of the time
    Team One 10 v 10 BA Team Two
    Veteran, ============== noob
    Longtimer,==============Forum Warrior
    Too Good,============== noob
    Veteran Plat 1,============== noob
    Veteran Plat 2,============== OP member
    Veteran Plat 3,============== noob
    Filthy Casual,============== noob
    noob,============== noob
    Level 1 noob ============== noob
    Crusher============== noob

    How hard is it to run a secondary matchmaking algorithm that re orders the team members to be more closely balanced.

    It feels like the matchmkaer just takes whatever players and randomly sprawls them across each team

    There are many ways to do this in regards to sorting list across two arrays (one for each team)

    One way is to compare the summation of MMR or “hidden” MMR on each team. Its quite obvious when you see a team of all high skilled, veteran, players against new players.

    how you would do this is take each pre matched team (mind you this would happen once all 20 players are found in brawl). Sum either player level, or MMR across each team. Subtract the difference. Then, run an iterative loop that re orders players till the difference converges.

    With the magic of computation. this would not have any affect on que times- as the initial matchmaking sequence is unchanged. This suggestion is only after all players are found- As a secondary match balancing algorithm that shouldnt take but seconds to sort

    #2125251

    jeacom512
    Participant

    Algorithm suggestion:

    sort players in a list by “hiden mmr”.
    select odd indexes to a team.
    select even indexes to another team.
    if the difference still too big:
    switch the players of the teams till be acceptable.

    but actually the current matchmaker seems to take in account the weapon types, just look when there are two teslas, aways will be one for each team never both in the same team.

    • This reply was modified 3 months, 1 week ago by  jeacom512.
    #2125256

    DrDoofenshmirtz
    Participant

    brawl doesnt use mmr or any matching at all once upon a time the game at least tried to give both teams even amounts of flyers but it was not possible for the game to know that certain bots could fly and often drones was classified as ground ……. and often hovers had rudders and was classified as fliers lol

    pretty much brawl is for mostly casual and very one sided always much like all matching used to be although the matcher isnt great it could be provided we had more players but for brawl and elimination it will never help because these modes simply dont even have hidden mmr

    #2125258

    TwiceAHuman
    Participant

    …One huge leap for mankind.

    #2125276

    brawl doesnt use mmr or any matching at all once upon a time the game at least tried to give both teams even amounts of flyers but it was not possible for the game to know that certain bots could fly and often drones was classified as ground ……. and often hovers had rudders and was classified as fliers lol
    pretty much brawl is for mostly casual and very one sided always much like all matching used to be although the matcher isnt great it could be provided we had more players but for brawl and elimination it will never help because these modes simply dont even have hidden mmr

    The point is to apply it to all game modes. What’s the point of accepting stupid un-balanced games. When (regardless if the initial match making is based on mmr or not) the games could have this one additional step to at least make game quite a bit more balanced. And is simply re-sorting a set of players. It’d be nice if you could volunteer a party to allowed to be re-sorted (just as long as your playing together) and of course it’s be a choice, not a requirement.

    Algorithm suggestion:
    sort players in a list by “hiden mmr”. select odd indexes to a team. select even indexes to another team. if the difference still too big: switch the players of the teams till be acceptable.
    but actually the current matchmaker seems to take in account the weapon types, just look when there are two teslas, aways will be one for each team never both in the same team.

    Doing a pure even – odd indexed set results in an overall difference that is the average difference between each indexed mmr, multiplied by the number of players on each team.

    A slightly better way is to do the same sorting then do {1}(2)(3){4}{5}(6)(7){8}{9}(10) ssuming the list has a somewhat cosistent gradient of MMR to choose from. The difference for this set about 1,;where a pure even to odd is 5. But of course that’s an even distribution assumption.

    Either way it’s good you see the point 🙂

    #2125291
    Nullpersona
    Nullpersona
    Participant

    Crusher

    I lol’d.

    Cheers.

    #2125299

    Crusher

    I lol’d.
    Cheers.

    If I wasn’t on a bus I would have fallen off my seat

    Honestly… Can’t we use “amount of CPU used on:” to determine weapons? That would be a start…

    #2125301

    Can’t we use “amount of CPU used on:” to determine weapons?

    Weapons and Modules. I feel this is a good idea. It can be used to sort who spent how much CPU into what part and balance the two teams’ loadout accordingly.

    However. Big however.

    However, I do acknowledge the fact that RC have so much possible weapons and modules (that being weapon types and tiers, and what modules) combination for just 5 loadout slots. Its not like the loadout system is limited to “primary weapon, secondary, tertiary, module 1 and 2” sets which can be balanced somewhat, like hero classes. Robocraft Loadout system is more free and that could make balancing longer to do. I say longer to do, not harder. The algorithm will always find a way to balance the teams, even if it means making compromises because of the lack of players. This means the matchmaker will be working longer because its taking more things into consideration. I’m fine with THAT. Not everyone would, though. If longer wait times means better quality matches, I’ll put up with longer queue times.

    And. Big and. But also with a big but.

    And this system could also be used to sort bot types, that being ground units and air units, along with the previous weapon balancing algorithm. No more 5 air vs 5 ground with no AA weapons, right? Not exactly. With RC’s abundance of freedom in bot making, what movement parts and what combination of them you use are probably unlimited (though not all of them viable). But unlike weapon loadout slots, movement parts are only limited by CPU. You can use any combination of movement parts in your inventory, viability notwithstanding. One way I can see this being balanced is categorizing movement parts into either ground based, air based, or supplementary movement type. The new MM algorithm see how much CPU is spent on what and decides the bot type depending on the majority. This would, again, put more work on the MM, making it match, well, matches, longer.

    This is a problem on both the weapon side and movement side. Someone please pick up on this and find a solution, because I smacked my face on a deadend.

    Note that this is written assuming all movement parts and weapon types are balanced and competitively viable. We’re not even there yet…

    #2125302

    Crusher

    I lol’d. Cheers.

    If I wasn’t on a bus I would have fallen off my seat
    Honestly… Can’t we use “amount of CPU used on:” to determine weapons? That would be a start…

    Do you mean the robobus? lol i still have that bot tho it sucks now 🙁
    whatever just make sure it is like this

    high–high
    high–high
    mid–mid
    mid–mid
    low-low
    low-low

    and not like this

    high–mid
    high–mid
    high-high
    low-low
    low-mid
    low-mid

    even tho both teams have the same average mmr the distribution is horrible in the second

    #2125346

    Do you mean the robobus? lol i still have that bot tho it sucks now

    I still have it too kek. Waiting on the 16 new colors to come in from the old days so I can repaint it, but no. I take a bus that goes between where I live and go to college (which is about an hour ride) This… Is why I’m always on the forums because I have literally nothing else to do xd

    However, I do acknowledge the fact that RC have so much possible weapons and modules (that being weapon types and tiers, and what modules) combination for just 5 loadout slots. Its not like the loadout system is limited to “primary weapon, secondary, tertiary, module 1 and 2” sets which can be balanced somewhat, like hero classes. Robocraft Loadout system is more free and that could make balancing longer to do. I say longer to do, not harder. The algorithm will always find a way to balance the teams, even if it means making compromises because of the lack of players. This means the matchmaker will be working longer because its taking more things into consideration. I’m fine with THAT. Not everyone would, though. If longer wait times means better quality matches, I’ll put up with longer queue times.

    But my general point is that typically… Your primary weapons are going to take up the most CPU than anything other weapons you would mount… Sticking a single legendary on your bot would mean you would have to spend more than 250-325 CPU on a different weapon type… At which point your probably going to use those more than the big gun.

    It would either way… Likely be far better off than all this MMR be. You could use it to classify movement parts too yis. But the only issue is classing thrusters…

    #2125350

    This assumes there are enough high skilled players to face off against high skill players at any point in time. If there are 100 players at any moment and about 5% of them have quite high skill compared to average, then how can matchmaker fairly matchmake. That’s not even taking the imbalance from being able to build at any CPU into account.

    5% is a randomly plucked number, but even if we say up it to 20% of 100 players are high skilled compared to average. It doesn’t take into account that maybe not all of that 20% are ready to go to battle at that moment.

    It’s not exactly fair to say just count up MMR and average it reagrdless. Because any GOOD player knows when it comes to newbs, noobs and nubs. Even an army of 5% accuracy nubs ain’t gonna stop ya one Rambo. I’ve seen single players destroy entire teams in all sorts of games.

    The bigger the pool of players, the better the matchmaker can work, even one as inaccurate as MMR. If the pool of player keeps reducing, there isn’t even an algorithm that could fix it.

    #2125351

    jeacom512
    Participant

    @all_my_friends_are_meme, we all know thats impossible to create a perfect fair game, what is needed is to reduce the likelyhood of unplesant maches.

    We dont need to minimize the differences, we need to maxmize the fun. Its not about querying perfectly, its about not leting a team with all pros vs a team of noobs.

    #2125355

    This assumes there are enough high skilled players to face off against high skill players at any point in time. If there are 100 players at any moment and about 5% of them have quite high skill compared to average, then how can matchmaker fairly matchmake. That’s not even taking the imbalance from being able to build at any CPU into account.
    5% is a randomly plucked number, but even if we say up it to 20% of 100 players are high skilled compared to average. It doesn’t take into account that maybe not all of that 20% are ready to go to battle at that moment.
    It’s not exactly fair to say just count up MMR and average it reagrdless. Because any GOOD player knows when it comes to newbs, noobs and nubs. Even an army of 5% accuracy nubs ain’t gonna stop ya one Rambo. I’ve seen single players destroy entire teams in all sorts of games.
    The bigger the pool of players, the better the matchmaker can work, even one as inaccurate as MMR. If the pool of player keeps reducing, there isn’t even an algorithm that could fix it.

    No.
    It doesn’t assume anything.

    It’s just an algorithm to re order teams after intial match making (which is next to non existent anyway)

    Currently there are too many one sided battle’s. And it’s completely obvious after playing this for year’s. It’s not hard to recognize players when you’ve been around this long. And it’s equally obvious when you see the match maker special throw all these long timers on one side. Then all noobs on the other side.

    @all_my_friends_are_meme, we all know thats impossible to create a perfect fair game, what is needed is to reduce the likelyhood of unplesant maches.
    We dont need to minimize the differences, we need to maxmize the fun. Its not about querying perfectly, its about not leting a team with all pros vs a team of noobs.

    Exactly.

    All those “Rambo’s” are usually on one side and then the rest on the other team.

    #2125408

    If this doesnt sum up my point i dont know what will.

    look at all those clan members on one team…and mayeb 3 on ours? let alon numerous skilled players on one side and not even a part or anything on ours to make up for atlas core party *
    *this is nothing against atlas core party in this example.

    THIS should be top priority.

    Attachments:
    You must be logged in to view attached files.
    #2125496

    c0l0ssus
    Participant

    Rewrite the entire MMR code to give players with high scores more MMR, despite winning or losing recent games. MMR should exist in ALL game modes. It shouldn’t be hard, but you already know what FJ is doing.

Viewing 15 posts - 1 through 15 (of 15 total)

You must be logged in to reply to this topic.