For other readers, please know that the 2000-3000 skill games and especially the 3000-4000 skill games graphs look funny because the amount of data is small in those skill bracket. I would argue that the 3000-4000 skill bracket is completely useless because the sample size is so small.
75% of games are between 8-26 minutes long. Long games like an hour long have less data also, especially at the higher skill brackets.
Where is this data taken from? For example the 3-4k graph. Is that from games with players with hive scores of 3-4k? Or is it gathered from any player with 3-4k hive skill in any game that they played including mixed hive scores?
Where is this data taken from? For example the 3-4k graph. Is that from games with players with hive scores of 3-4k? Or is it gathered from any player with 3-4k hive skill in any game that they played including mixed hive scores?
The source of the data was shown in the OP. Wonitor collects data on NS2. This post in another thread about wonitor has me describing the data in a little more detail.
It would be great to see how many games have been recorded in each bracket. I feel like a lot of these bars are skewed by low sample sizes.
The 3000-4000 bracket is skewed from low sample size, and so are the data on long games in 2000+ hive skill games.
The following graph represents a small section of data of which I added constraints to try to better show what players might actually see right now. This is the distribution of the server average hive skill.
I don't think anyone in this thread, myself in included, is posting graphs that have similar constraints. Some of the wonitor data or so has balance mods which further limit its usefulness. You can cut out all that, but that further reduces your sample size and takes more effort. I admit I am being lazy myself even though I know better.
Where is this data taken from? For example the 3-4k graph. Is that from games with players with hive scores of 3-4k? Or is it gathered from any player with 3-4k hive skill in any game that they played including mixed hive scores?
I have seen you being very active in these forums, which is actually great. It has been about the same 30-50 same people posting here for years now. We seem to keep coming back to roughly the same 10-12 topics or so every few months, with a slight twist or a complete rehash of the same stuff talked about last here. A new voice is a very good thing. What I am about to say seems potentially elitist, but I really don't mean it that way. There is a wealth of context from some of these other threads that you may be missing. You seem very interested, and you may enjoy reading some old threads to gain this context. I really don't mean this in an elitist way. This is something I do in forums I am new too, so I can sort of catch up on the conversation.
If we are being honest with ourselves, most of us come here to do pointless argumentation and discussion on a game where our ideas will probably never go anywhere. We are all armchair game designers here. We seem to like the process more than the results it seems.
Where is this data taken from? For example the 3-4k graph. Is that from games with players with hive scores of 3-4k? Or is it gathered from any player with 3-4k hive skill in any game that they played including mixed hive scores?
The source of the data was shown in the OP. Wonitor collects data on NS2. This post in another thread about wonitor has me describing the data in a little more detail.
It would be great to see how many games have been recorded in each bracket. I feel like a lot of these bars are skewed by low sample sizes.
The 3000-4000 bracket is skewed from low sample size, and so are the data on long games in 2000+ hive skill games.
The following graph represents a small section of data of which I added constraints to try to better show what players might actually see right now. This is the distribution of the server average hive skill.
I don't think anyone in this thread, myself in included, is posting graphs that have similar constraints. Some of the wonitor data or so has balance mods which further limit its usefulness. You can cut out all that, but that further reduces your sample size and takes more effort. I admit I am being lazy myself even though I know better.
Where is this data taken from? For example the 3-4k graph. Is that from games with players with hive scores of 3-4k? Or is it gathered from any player with 3-4k hive skill in any game that they played including mixed hive scores?
I have seen you being very active in these forums, which is actually great. It has been about the same 30-50 same people posting here for years now. We seem to keep coming back to roughly the same 10-12 topics or so every few months, with a slight twist or a complete rehash of the same stuff talked about last here. A new voice is a very good thing. What I am about to say seems potentially elitist, but I really don't mean it that way. There is a wealth of context from some of these other threads that you may be missing. You seem very interested, and you may enjoy reading some old threads to gain this context. I really don't mean this in an elitist way. This is something I do in forums I am new too, so I can sort of catch up on the conversation.
If we are being honest with ourselves, most of us come here to do pointless argumentation and discussion on a game where our ideas will probably never go anywhere. We are all armchair game designers here. We seem to like the process more than the results it seems.
For other readers, please know that the 2000-3000 skill games and especially the 3000-4000 skill games graphs look funny because the amount of data is small in those skill bracket. I would argue that the 3000-4000 skill bracket is completely useless because the sample size is so small.
75% of games are between 8-26 minutes long. Long games like an hour long have less data also, especially at the higher skill brackets.
@Nordic
I wonder if the skill braket above 2k is smaller because the higherskilled guys in those teams tend to leave at the end of the round? Because the other lowskilled guys tend to drag a lost game out to the bitter end? ... Can you scrutinise that the data is not flawed by that issue?
If you cannot than that would shift the graph into the right direction.. that there are more higher skilled games going on than "recorded"
@Mouse I think for those graphs it would be better if you opened a braket starting at 1,5k is like 50% improvement... compared to 1kl
@Handschuh that is possibly a variable. Given that players over 2000 hive skill are a very minority group overall, I think it is more likely that there are just that few of those games. I was posting stats from all wonitor data available which includes all balance mods, woozas server, and games played years ago. If we wanted really relevant statistics we would remove those variables, but I dont want to put to much time into this right now, and I doubt the others will either.
I have been meaning to post some high quality statistics for awhile now but I keep putting it off. I really should find a way to automate most of it.
Monthly "State of NS2" graphs and stats would be hilariously awesome
I have wanted to do it every 6 months. Monthly is far too often in terms of work load, and nothing would really change much month to month. I have some really awesome graphs I want to make, but it is a struggle to get me going. The only reason I hang out here still is because I plan to share those graphs.
MephillesGermanyJoin Date: 2013-08-07Member: 186634Members, NS2 Map Tester, NS2 Community Developer
edited March 2018
what kind of graphs?
EDIT: Not sure if asked already but at what point does hive predict the outcome of the round? I had a game yesterday where aliens started with like 100 skill less but like 5 minutes before the aliens lost someone joined and they were 300 points ahead. I was wondering if at the end of round hive thought that aliens should have won that.
@Handschuh that is possibly a variable. Given that players over 2000 hive skill are a very minority group overall, I think it is more likely that there are just that few of those games. I was posting stats from all wonitor data available which includes all balance mods, woozas server, and games played years ago. If we wanted really relevant statistics we would remove those variables, but I dont want to put to much time into this right now, and I doubt the others will either.
I have been meaning to post some high quality statistics for awhile now but I keep putting it off. I really should find a way to automate most of it.
In my opinion every statistic shown lost all relevance withany data as it doesn't resolve the issue that it counts only the "remining players" when the round is over.
And mostly it's the best guys of the team which leave, since they're going to be afk (not wasting time)... or don't have fun in turtling much.
If I'm spec to wait for a slot it's right before "conceding" when you can snatch one, because they leave the game. Same issue is that right before the end of the round they're pressing F4... in the hope the round ends more quickly that way
This happens on a regular basis almost all day, maybe not every game and maybe not on every server, but as long as I see that on a regular basis - it screws every statistic.
My claim is that the average of the loosing team ends not very seldom right before the end drastically...
MephillesGermanyJoin Date: 2013-08-07Member: 186634Members, NS2 Map Tester, NS2 Community Developer
Not all data lost it's revelancy. When it comes to simple things like which side won how much then it doesn't matter if the strong people left at the end. The only data that is skewed with that are the data that are plotting hive skill or use it as constraint. But a lot of data found on wonitor is still valid independant from the hive skill
Undoubtedly, higher skilled people do tend to leave before the end. No one can say with certainty how much of a factor that is. That does not make all of the data irrelevant, but it does limit the validity of the stats in the OP.
There are four NS2 datasets. There is the hive dataset of which only UWE and myself have. This includes player data and some limited game data. There is the sponitor dataset which only UWE and I have. Sponitor has a very similar data to wonitor, and has the same problems, and more. Wonitor is based on sponitor. There is the NS2+ dataset which has a ton of data but is difficult to work with or derive meaning from. Then we have the wonitor dataset, which is the best dataset we have on NS2 games, which does not mean it is without fault.
How valid the statistics produced from that data are, is dependent on the statistics created. The statistics in the OP, while interesting, do lose some validity. The amount of validity lost really comes down to a matter of opinion. Does that opinion even matter? Not really because these statistics here aren't even useful. They are fun and interesting, but not useful. It is what it is.
This does not make every statistic irrelevant. Statistics by definition do not tell the whole story.
Data and statistics on high skill games has always been questionable. Not only for the reasons you are stating, but from a severe lack of data. Even with people leaving games, it does not account for the sheer lack of games at high skill levels. We have just enough data when using all historical data to even scrape something reasonably useful out of 2000+ skilled games. 3000+ skilled games do not even have enough data recorded to do anything worthwhile with them even with all games recorded historically. I also usually cut out comp games because they use a different balance than vanilla, which further reduces the sample of high skill games. The OP makes no such distinction based on balance mods.
Mods, ns2+options, lerk movements, fade attacking.... all the other annoying things that take advantage of... issues.
You want fair games? NS2 without ns2+ and without mods would help.
It would only be unfair if you couldnt make the same changes as the player you are complaining about. You have the same opportunity to change the HUD, or even hide it, change the alien vision, hitregs, etc...
Mods, ns2+options, lerk movements, fade attacking.... all the other annoying things that take advantage of... issues.
You want fair games? NS2 without ns2+ and without mods would help.
It would only be unfair if you couldnt make the same changes as the player you are complaining about. You have the same opportunity to change the HUD, or even hide it, change the alien vision, hitregs, etc...
Ignorance is not an excuse :P
Removing models to give yourself a larger view before it could be found in ns2+ options was just plain cheating IMO.
You can go on and think new players are ignorant, but I for one was surprised to see such an option in a fps.
Removing models to give yourself a larger view before it could be found in ns2+ options was just plain cheating IMO.
You can go on and think new players are ignorant, but I for one was surprised to see such an option in a fps.
Just playing the devils advocate here: Is having a better hardware cheating? Is taking stimulants cheating?
They both affect a player's performance...
MephillesGermanyJoin Date: 2013-08-07Member: 186634Members, NS2 Map Tester, NS2 Community Developer
I agree on the fact that ns2+ is a bit unintuitive since the option don't appear in the main menu unless you have the mod mounted. But afaik the devs want to inegrate the ns2+ settings into the vanilla options. It's just not on top of the priority list atm.
Removing models to give yourself a larger view before it could be found in ns2+ options was just plain cheating IMO.
You can go on and think new players are ignorant, but I for one was surprised to see such an option in a fps.
Just playing the devils advocate here: Is having a better hardware cheating? Is taking stimulants cheating?
They both affect a player's performance...
I understand the aim of your question... What is cheating and what is unfair are two different angles. One is obtuse compaired to the other.
Caffine is hell of a drug... I doubt anyone wants to see a league tourney without it. From bawls to grape nos...
Obviously someone with the advantage has one, but I do not believe it is what people generally consider an unfair advantage.
The poll here on these forums on increasing the allowed hertz for monitors is sound thinking in my opinion.
I for one only have a 60hertz monitor. Maybe, me buying a 120hert will bring my 2k to a 3.5k... I doubt it, but maybe I will be solo'ing more lerks and landing more hits on fades... If so, it might feel unfair by comparison... I'd have to go buy one and let you know.
Mods, ns2+options, lerk movements, fade attacking.... all the other annoying things that take advantage of... issues.
You want fair games? NS2 without ns2+ and without mods would help.
It would only be unfair if you couldnt make the same changes as the player you are complaining about. You have the same opportunity to change the HUD, or even hide it, change the alien vision, hitregs, etc...
Ignorance is not an excuse :P
Yeah. People always ask how I kill them in two bites, so I just tell them to check their NS2+ options
Which means that 4k player is not the sum of two 2k players but their multiplication...?
I didn't mean to say what I said earlier. I meant to make a quick post, but it was not exactly a correct post.
What I meant to say is that hive skill has a log normal distribution. "Normal distribution arises when the thing you are measuring is a sum of independent factors. Log normal arises when it's a product of independent factors"
A log normal distribution for all players is very different than hive skill operating on a logarithmic scale. It would have been better if I had not made that comment at all.
@moultano could probably say more though, and a lot more accurately than I can.
Which means that 4k player is not the sum of two 2k players but their multiplication...?
Both are true in a sense. The hive model assumes that if you start with even teams, and add two 2k players to one team and a rookie and a 4k player to the other team the teams are even.
It's also true that the way that skill differences affect the win probability is multiplicative. You can rewrite the model to say "the expected win/loss ratio of this matchup is exp(avg_skill_difference/200)" in which case adding to the skill has the effect of multiplying the win-loss ratio.
For the purposes of teams of different sizes, the model assumes that being down a player is functionally equivalent to having a rookie in that slot.
Comments
75% of games are between 8-26 minutes long. Long games like an hour long have less data also, especially at the higher skill brackets.
[EDIT] There's only been 59 games where the average skill level is greater than 4000
The 3000-4000 bracket is skewed from low sample size, and so are the data on long games in 2000+ hive skill games.
The following graph represents a small section of data of which I added constraints to try to better show what players might actually see right now. This is the distribution of the server average hive skill.
I don't think anyone in this thread, myself in included, is posting graphs that have similar constraints. Some of the wonitor data or so has balance mods which further limit its usefulness. You can cut out all that, but that further reduces your sample size and takes more effort. I admit I am being lazy myself even though I know better.
I have seen you being very active in these forums, which is actually great. It has been about the same 30-50 same people posting here for years now. We seem to keep coming back to roughly the same 10-12 topics or so every few months, with a slight twist or a complete rehash of the same stuff talked about last here. A new voice is a very good thing. What I am about to say seems potentially elitist, but I really don't mean it that way. There is a wealth of context from some of these other threads that you may be missing. You seem very interested, and you may enjoy reading some old threads to gain this context. I really don't mean this in an elitist way. This is something I do in forums I am new too, so I can sort of catch up on the conversation.
If we are being honest with ourselves, most of us come here to do pointless argumentation and discussion on a game where our ideas will probably never go anywhere. We are all armchair game designers here. We seem to like the process more than the results it seems.
God what an elitist.
I wonder if the skill braket above 2k is smaller because the higherskilled guys in those teams tend to leave at the end of the round? Because the other lowskilled guys tend to drag a lost game out to the bitter end? ... Can you scrutinise that the data is not flawed by that issue?
If you cannot than that would shift the graph into the right direction.. that there are more higher skilled games going on than "recorded"
@Mouse I think for those graphs it would be better if you opened a braket starting at 1,5k is like 50% improvement... compared to 1kl
Skill level 0+ (grouped every 500)
Skill level 1000-2000 (grouped every 100)
I have been meaning to post some high quality statistics for awhile now but I keep putting it off. I really should find a way to automate most of it.
I have wanted to do it every 6 months. Monthly is far too often in terms of work load, and nothing would really change much month to month. I have some really awesome graphs I want to make, but it is a struggle to get me going. The only reason I hang out here still is because I plan to share those graphs.
EDIT: Not sure if asked already but at what point does hive predict the outcome of the round? I had a game yesterday where aliens started with like 100 skill less but like 5 minutes before the aliens lost someone joined and they were 300 points ahead. I was wondering if at the end of round hive thought that aliens should have won that.
In my opinion every statistic shown lost all relevance withany data as it doesn't resolve the issue that it counts only the "remining players" when the round is over.
And mostly it's the best guys of the team which leave, since they're going to be afk (not wasting time)... or don't have fun in turtling much.
If I'm spec to wait for a slot it's right before "conceding" when you can snatch one, because they leave the game. Same issue is that right before the end of the round they're pressing F4... in the hope the round ends more quickly that way
This happens on a regular basis almost all day, maybe not every game and maybe not on every server, but as long as I see that on a regular basis - it screws every statistic.
My claim is that the average of the loosing team ends not very seldom right before the end drastically...
There are four NS2 datasets. There is the hive dataset of which only UWE and myself have. This includes player data and some limited game data. There is the sponitor dataset which only UWE and I have. Sponitor has a very similar data to wonitor, and has the same problems, and more. Wonitor is based on sponitor. There is the NS2+ dataset which has a ton of data but is difficult to work with or derive meaning from. Then we have the wonitor dataset, which is the best dataset we have on NS2 games, which does not mean it is without fault.
Wonitor has the following data within it:
How valid the statistics produced from that data are, is dependent on the statistics created. The statistics in the OP, while interesting, do lose some validity. The amount of validity lost really comes down to a matter of opinion. Does that opinion even matter? Not really because these statistics here aren't even useful. They are fun and interesting, but not useful. It is what it is.
This does not make every statistic irrelevant. Statistics by definition do not tell the whole story.
Data and statistics on high skill games has always been questionable. Not only for the reasons you are stating, but from a severe lack of data. Even with people leaving games, it does not account for the sheer lack of games at high skill levels. We have just enough data when using all historical data to even scrape something reasonably useful out of 2000+ skilled games. 3000+ skilled games do not even have enough data recorded to do anything worthwhile with them even with all games recorded historically. I also usually cut out comp games because they use a different balance than vanilla, which further reduces the sample of high skill games. The OP makes no such distinction based on balance mods.
I've seen great shooting/lerk/fade from 1.8k and up.
From 3-4k well... I'll let you all think about it.
In pubs, a player that wins more than he losses... maybe isn't playing fair.
Or maybe he's the only player in the server that knows heads from tails in this game...
Mods, ns2+options, lerk movements, fade attacking.... all the other annoying things that take advantage of... issues.
You want fair games? NS2 without ns2+ and without mods would help.
Some would call it a skill gap... I think it's a level of lameness that has to be adjusted for.
It would only be unfair if you couldnt make the same changes as the player you are complaining about. You have the same opportunity to change the HUD, or even hide it, change the alien vision, hitregs, etc...
Ignorance is not an excuse :P
Removing models to give yourself a larger view before it could be found in ns2+ options was just plain cheating IMO.
You can go on and think new players are ignorant, but I for one was surprised to see such an option in a fps.
Just playing the devils advocate here: Is having a better hardware cheating? Is taking stimulants cheating?
They both affect a player's performance...
Edit: How I said this is not right. I quickly said this earlier today and it just came out wrong. I will make a new post below.
I understand the aim of your question... What is cheating and what is unfair are two different angles. One is obtuse compaired to the other.
Caffine is hell of a drug... I doubt anyone wants to see a league tourney without it. From bawls to grape nos...
Obviously someone with the advantage has one, but I do not believe it is what people generally consider an unfair advantage.
The poll here on these forums on increasing the allowed hertz for monitors is sound thinking in my opinion.
I for one only have a 60hertz monitor. Maybe, me buying a 120hert will bring my 2k to a 3.5k... I doubt it, but maybe I will be solo'ing more lerks and landing more hits on fades... If so, it might feel unfair by comparison... I'd have to go buy one and let you know.
Yeah. People always ask how I kill them in two bites, so I just tell them to check their NS2+ options
I didn't mean to say what I said earlier. I meant to make a quick post, but it was not exactly a correct post.
What I meant to say is that hive skill has a log normal distribution. "Normal distribution arises when the thing you are measuring is a sum of independent factors. Log normal arises when it's a product of independent factors"
A log normal distribution for all players is very different than hive skill operating on a logarithmic scale. It would have been better if I had not made that comment at all.
@moultano could probably say more though, and a lot more accurately than I can.
Both are true in a sense. The hive model assumes that if you start with even teams, and add two 2k players to one team and a rookie and a 4k player to the other team the teams are even.
It's also true that the way that skill differences affect the win probability is multiplicative. You can rewrite the model to say "the expected win/loss ratio of this matchup is exp(avg_skill_difference/200)" in which case adding to the skill has the effect of multiplying the win-loss ratio.
For the purposes of teams of different sizes, the model assumes that being down a player is functionally equivalent to having a rookie in that slot.