!predict gives weird, obviously incorrect results in small teams

sort
libcrusher 5 years ago (edited 5 years ago) Earlier today, in the lobby for the small teams game, !predict suggested that I had a ~51% chance of victory when facing two opponents on a small map. Both opponents had significantly more experience than I did. One had a rank color of top 5% (blue) and the other of top 60% (orange). I, as you can see, have a rank color of top 40%. The blue player alone should have been heavily favored against me, let alone with a teammate. Obviously, this is concerning for matchmaking if nothing else (it wanted to sort me into a 2v1 match against the other two players). Any ideas about this? I took a screenshot but it turned out just of my desktop background, not Zero-K, so no screen, sorry. (I don't know why that happened either.) +0 / -0
Manu12 5 years ago (edited 5 years ago) You underestimate the impact of a low elo player in team games. In this case the orange player was obviously placed in the team with the blue player to counter his higher elo. Predict works as (Total Elo)/(number of players) and compares this number of both teams. While he got a "teammate" you are generally better then the lower player and have two commanders to compensate the fact that it is a 2v1. This is why predict showed up as ~50% winchance. Do not forget that the higher elo player in the team only gets 50% of the metal income because he has a teammate. +2 / -0
libcrusher 5 years ago Interesting! That seems like a subpar balancing formula insofar as it fails to account for the advantage of having two human brains (and therefore two attention streams, roughly doubled APM, etc.) rather than one. I wonder if it could be tweaked to account for this fact. Certainly the actual win rate in my example situation would be ~an order of magnitude off what was predicted. +0 / -0
CatLady 5 years ago (edited 5 years ago) If anything, it could add WHR of all players without dividing between number of players in case of uneven teams... --- +0 / -0
Sortale 5 years ago libcrusher absent from all the rest, what kind of pairing would have been superior in that particular case of yours? I mean for 1 match between the top 5% and top 40%, the 60% player would act as a limiter on resource receive. You are unhappy with 60% being paired with the top 5%, you would hate it if 60% is paired with you. Do you want him alone against both you and the 5% player? that would be an even worse balance. in this case the balance is just trying to do the best in a bad situation. +0 / -0
the_green_squig 5 years ago Manu12 would probably beat two of me if he had two comms (probably even 1 tbh) to play with but I do appreciate why the prediction often looks weird. I sometimes wonder at what WHR difference 1 elite player beats 2 merely good ones. +3 / -0
CrazyEddie 5 years ago Let's find out! Someone organize a 2v1 tournament and see what happens! +5 / -0
Steel_Blue 5 years ago When I play 2v1, I think my win chance is baised more on my enemy's failures than my own success. If they do not properly expand over their half of the map, I'll just shove units down one of their throats until its's a 1v1, I should have more eco than a single one of them and I almost expect them to play for themselves. If one of them doesn't raid me, that's just playing a 1v1 where I have 2 coms, so I can play normal attack/defence. If they both raid, things get tricky. I love micro when I play, I try my best never to use fightmove/attackmove because I feel I can control my units better than the AI, but when I'm fighting on 3 fights at once against 2 mouses I can't just out micro my opponent on every front. If the enemy expands well and raids a little, it will become impossible for me to fight across the 50/50 even map control. If I think the enemy will hold their 50% of the map, I will try to focus my army and break through the front and cause chaos in their back lines, because the enemy's economy will be split 50/50 which means a counter attack will not be as strong. Also, people don't base trade when I do this, so my economy's guaranteed to keep strength if I divert their attention to their map half. +1 / -0
dyth68 5 years ago Sortale: You're confusing getting accurate !predict numbers and optimal team pairings. libcrusher vs the other two might indeed have been the most balanced choice, but it might still be just a 20% chance of winning. the_green_squig : In my opinion, the extra APM and attention from decent players will allow them to stomp even good ones on large maps. I don't think Godde or Zenfur could take on three silver players. +0 / -0
Manu12 5 years ago The most important factor in 1v2 - 1v3 is time. If your opponent does constantly harass you and is diverting your attention it is somewhat hard to build up eco and form a counterattack. But if you are left alone for a small time window you can even 1v3 with only 1 commander. e.g. https://zero-k.info/Battles/Detail/741269 +2 / -0

Forum index > General discussion >