Hive skill difference vs winratio
pebl
Join Date: 2016-02-09 Member: 212816Members
Join Date: 2016-02-09 Member: 212816Members
I was accused of pulling number out of my ass yesterday when I mention right at the start of game the likelihood of marines winning.
I was not aware of how difficult it was to copy urls into the game, so I decided to make a post about it instead.
First here is a link of how it theoretically should look like (stolen from @Nordic ) in wolfram.
Here is some wonitor urls that should show the same curve (win ratio) vs hive skill difference:
House of Awesome!
Thirsty Onos
DiamondGamers
Survival of the Fattest
NS2Large
I would say they match the curve ok.
So in the grand scheme; a hive skill difference really matters (no matter what ppl claim), and therefor your hive skill really tell something about you (no matter what a lot of ppl will claim),
and perhaps splitting hive skill in alien/marines may not be that important (too few where it matters/or they underestimate their value on their bad team).
And most important - do reshuffle if the hive skill difference is large, even if the game already have begone.
And a plea - if you are that elite that you change the hive skill difference significant - dont join a game midtgame.
Note that
1) this tells nothing about how good the shuffle heuristic is (at all).
2) Yes the graphs are messy in the extremes.
3) That is why I care about hive skill.
I was not aware of how difficult it was to copy urls into the game, so I decided to make a post about it instead.
First here is a link of how it theoretically should look like (stolen from @Nordic ) in wolfram.
Here is some wonitor urls that should show the same curve (win ratio) vs hive skill difference:
House of Awesome!
Thirsty Onos
DiamondGamers
Survival of the Fattest
NS2Large
I would say they match the curve ok.
So in the grand scheme; a hive skill difference really matters (no matter what ppl claim), and therefor your hive skill really tell something about you (no matter what a lot of ppl will claim),
and perhaps splitting hive skill in alien/marines may not be that important (too few where it matters/or they underestimate their value on their bad team).
And most important - do reshuffle if the hive skill difference is large, even if the game already have begone.
And a plea - if you are that elite that you change the hive skill difference significant - dont join a game midtgame.
Note that
1) this tells nothing about how good the shuffle heuristic is (at all).
2) Yes the graphs are messy in the extremes.
3) That is why I care about hive skill.
Comments
If you're going to request this, then you damn well better request the inverse too. Especially if you want to talk of fairness. If averages are in the mid two thousands (and they frequently are), then you had better make a similar plea that anyone in the low hundreds needs to not join midgame, either.
Dead weight can ruin a game just as easily.
I suspect this wouldn't go over too well with a lot of people who feel that high skilled players are the "problem", though.
edit: grammar/spelling
Like, I understand how a dead weight could influence a 6v6 competitive match. And I understand having a dead weight blocking a potential spot. But in a pub game where you already have made teams, and someone joins along midgame.. How is he making your chances *worse* than having nobody at all?
It's not like dota where a bad team mate will feed the opponent with gold and exp. If a team mate does poorly, at worst he clogs up the infantry portals a little bit or takes a few too many eggs.
I'm sorry, but I'm not going to list the multitude of reasons why and how a dead weight player can impact a game.
Frankly, I'm advocating fairness. If a higher skilled player than the average can't join, than neither should a lower skilled.
At the end of the day the two responses prove my point. Everyone's okay with screwing over a higher skilled player who wants to play, but not the lower skilled.
With the same data I added the constraint of games with an average 2000 or more hive skill. I was thinking the curve might break down at that skill level. This sample still has thousands of games in it, and more than any individual server does from the links in the OP. It still holds up fairly well even at the higher skill levels.
This makes me wonder how players would react if they were provided this information. What if NS2+ showed the average skill level and showed the theoretical chance of each team winning based on the function shared in the OP. I wouldn't think that would be programmatically hard to make. The hard part would be getting a modder to do it and to get UWE to accept the addition into NS2+ on github.
Ahh, I completely agree with you; the inverse is bad too. However the ppl reading this forum are not from that (hive skill) group.
Secondly I think you overestimates how many games are 2500+, so I seldom if ever see this inverse problem. Average skill games.
But I should have worded it differently.
I did try to find a good estimate but it is not that easy. So I made the second best; A hive skill end game graph in ns2+ instead: pull request, and hope ppl get a better idea of when 100 skill points matter on which servers and with X amount of players.
There could be some systematic over or underestimation depending on the size of the games. The model makes assumptions about how skills combine as the game size increases that could be off in practice, and off by a different amount for each game size. The model assumes that the effect of one good players is diluted as the games get bigger, and that all the games should get closer to a tossup as game size gets bigger. (An alternative that would give the predictions the same variance at every gamesize would divide the skill difference by the sqrt of the number of players instead of the number itself.)
In my opinion the Winrate of the aliens need to be higher than 50% with Hiveskill for example 2k vs 2k... even seperate hiveskills won't help much.
I think the biggest issue is that there quite some people that have higher hiveskill than they should've because they're winning games against much stronger enemies from time to time they shouldn't have via Tunnel/Baserush. (At any point during the game)
Especially on lowerskilled (1500-200) servers I notice those things... so If I'm alien with guys like those my only choice is to win with them via baserush or play lerk like a frigging lawnmower, because their alienplay is so incredibly bad... it's insane... but they know how to sneak through in order to get a tunnel and baserush...
I have wanted to know this for ages!
The issue is that it happens, and it happens while the teams are incredibly unbalanced... ;-)
So a number doesn't respresent the "hiveskillvalue"
I remember days on 8bit server with a baserushtunnel every game... several times... it only worked on every 3rd game... but still....
The attempt at getting even a rough estimate didn't even use hive skill. I have quoted the information.
It seems to be the default desperation strat.
1) Big skill values above 4000 entirely spoils field player's game experience if you playing hive shuffled match because it doesn't make difference between alien or marine or commander. So if you shuffled in a team with average skill below 2500 probably you have to carry them. You can't play commander or your team will suffer from lack of field skill (that's frustrating since often I want to play it). Also you there is a BIG difference between your possibility to carry your team by aliens and by marines (you can shoot plenty of aliens if you medspamed well but most likely you will face problems with killing more than 2 marines for a short period of time as alien).
In practice it leads to the low alien winrate for that kind of players and high marine winrate. It doesn't affect player's hive skill since player has 50% chance to play marines or aliens. But for example if you are an alien player with 1800 hive skill, server has average skill 2300 and you has one or two players in your team with skill above 3800 there is a very little possibility you will win that round (maybe ~20%).
2) Whole game has a big problem with marine winrate in low skilled matches (below 2000 I suppose) because marine gameplay is less forgiving mistakes while as alien player in low skilled match you can just play faceroll all the time and SUDDENLY marines lose everything they got. I don't think it's a crucial problem but could be solved too if devs will come up with a separated skill system.
We have graphs for that too. This is with all recorded data from wonitor. The average "average game skill" is going to be about 1400 for this graph.
That's an awesome graph, and really confirms that rookies are terrible aliens. I wonder if the likelihood of high skill games to have a high skill difference confounds things a little bit? If the skill differences for high skill games are more likely to be far from 0, then this might push the observed average to be closer to 0.5 than it would be naturally since the skill difference would dominate the race bias. Still pretty compelling chart either way. If it's possible you could plot avg(abs(skill difference)) vs avg skill.
Could you clean up the graphs by
A) only showing games where the skill difference only varied by a small percent throughout tge game
As far as I understand it, the skill differences recorded are how the game ended not how it began. So we have no context on if the skill changed throughout the game, only how it ended.
Iirc there's a trend that shows that aliens tend to win longer games. I wonder how the average game length changes as skill levels change; then whether there's a correlation between skill level, game length, and win rates.
e.g.: Games are shorter in low skill games, which leads to more marine wins.
^
Why does it calculate this way? The hive skill at the end of the round seems like the least relevant skill value for predicting wins. And it seems like players on the losing team more frequently quit before the end. (I'm not sure whether low or high elo players would have a greater tendency to quit before the end, but just given that there are many more of the former, I'd guess that late game rage quits tend to drag up the losing team's elo.)
Wonitor is designed to be performance friendly. Wonitor collects its data at the end of a round to not affect game performance during the round.
The deathmaps are for those servers who enable wonitor 2.0 which has additional features but also additional bandwidth and data requirements. Not every server operator is able or willing to record with wonitor 2.0.
@Brute made wonitor. He could answer these questions better than I could.
@Mephilles did some research: