New PSR System

[KGB_Killer]

I spent roughly 6 hours yesterday coming up with a balanced PSR system that better reflects a persons actual skill level as opposed to just winning and losing.

My system incorporates Kills per game, deaths per game, wins, losses, assist per game, creep kills per game, creep denies per game, average game length, and is set up to be based off of the last 30 games a person has played.

I have created a sample of 32 people with all of there stats and what there psr would be under my system; which i feel is an accurate number for thier skill.

The only thing i was unable to incorporate is the AWC in game, which stands for Average Winning Chance in game (this refers to the average percentage your team has been given prior to the games start in which u have played (the percentage thingy above your team before the game starts). This is included in my formula but i just set everyones to a default of 50%.

Unfortunitly it would be much too complicated for many people to understand the formula by which these results were calculated but I ask you just to vote based on whether you think the PSR in my chart is an accurate number. (im not being an ass but people will make arguments which have no basis if i post the numbers).

What i am willing to do is post what your personal PSR would be if this system were implimented.

Just post your current name on HoN and that you wish to recieve your PSR and I will do that.

Also if you must know the formula and you have microsoft excel i would be more then willing to send the spreadsheet to you. Just post.

As for now here is the link to the data.


http://i26.tinypic.com/eaqvdg.jpg

(Zoom in to get a good view)


http://hon.slawed.net/sig2.php?nick=KGB_Killer&img=DementedShaman&style=t&bg=000000&bd=DC143C&tx=FFFF00&va=FFFFFF&tx2=3ABCE5

[KGB_Killer]

If you vote please post a reply also, with a reason so i can improve the system.

[_Archangel_]

It's really hard to analyse the raw data and read your wall of text to try and understand it, let alone try deciding whether it's accurate or not. Perhaps try graphing the information :\

[Thugnificent]

The current PSR system is horrible. It in no way reflects skill. All it seems to do is scare of pubs and make it take 20 min just to fill a game. I'm tired of playing 5-10 games with my friends and getting 5 rating per win, then if I try to play a "balanced" game and lose, I lose 40.

[Blocky]

Looks ok, but wouldnt that make people who play support/gankers alittle underrated. Considering the low amount of Creepkills pergame, but then i assume the assists will reflect the diffrence. Either way i think its good the current one sucks because you cant always win a pub game no matter how good you play in a 5v5.

[Raiynewerth]

I like it.

Perhaps you could compare the A/D ratio to the K/D ratio and take the more favourable statistic to apply / increase the percentage of importance it holds to the PSR?

[````____````]

Looks ok, but wouldnt that make people who play support/gankers alittle underrated. Considering the low amount of Creepkills pergame, but then i assume the assists will reflect the diffrence. Either way i think its good the current one sucks because you cant always win a pub game no matter how good you play in a 5v5.

Agree..

You should try to find the average or expected assists of a support hero in one game, and the average creep kills in one game from the other heroes. This way, you can say: one assist is equal to x creep kills. This would likely balance the influence on rating from assists as support heroes won't get many kills.

Just a suggestion.

[Drunken_Jedi]

This just doesn't work.

HoN has exactly one objective and that is destroying the other team's main building. How you get there is completely irrelevant.
Kills, deaths, assists, creep kills, denies and all those other stats are ultimately irrelevant. They only matter because killing opposing heroes, staying alive and killing creeps usually increases your chances of winning the game. With your PSR system, those stats are no longer just means to an end; instead they become part of the actual objective.
This is very bad because it will make players even more reluctant to play support heroes and I really don't want to have to constantly play on teams with 3+ carries and semi carries.

The only reliable way of measuring a player's true skill is the way PSR currently works. The drawback is that this stat requires a lot of games to become reliable. For someone who has played 20 games, PSR just doesn't mean anything. Even with 100 games, PSR is only a very rough estimate of true skill, but it's the only stat that has no systematic bias towards a certain playstyle or hero preference.

[ma5]

The current PSR system is horrible. It in no way reflects skill. All it seems to do is scare of pubs and make it take 20 min just to fill a game. I'm tired of playing 5-10 games with my friends and getting 5 rating per win, then if I try to play a "balanced" game and lose, I lose 40.
Rofl! That means teh system is working!!!!

[gotlucky]

This just doesn't work.

HoN has exactly one objective and that is destroying the other team's main building. How you get there is completely irrelevant.
Kills, deaths, assists, creep kills, denies and all those other stats are ultimately irrelevant. They only matter because killing opposing heroes, staying alive and killing creeps usually increases your chances of winning the game. With your PSR system, those stats are no longer just means to an end; instead they become part of the actual objective.
This is very bad because it will make players even more reluctant to play support heroes and I really don't want to have to constantly play on teams with 3+ carries and semi carries.

The only reliable way of measuring a player's true skill is the way PSR currently works. The drawback is that this stat requires a lot of games to become reliable. For someone who has played 20 games, PSR just doesn't mean anything. Even with 100 games, PSR is only a very rough estimate of true skill, but it's the only stat that has no systematic bias towards a certain playstyle or hero preference.
Is PSR currently done by the ELO system? Is it just based on win/loss? If it is, I completely agree with you. I would like to point out that if it is win/loss, however, the only flaw with ELO is that it is meant for 1v1 and not team games. In my opinion ELO is still the best system out there for rating skill, though.

[Drunken_Jedi]

Is PSR currently done by the ELO system? Is it just based on win/loss? If it is, I completely agree with you. I would like to point out that if it is win/loss, however, the only flaw with ELO is that it is meant for 1v1 and not team games. In my opinion ELO is still the best system out there for rating skill, though.
It's either ELO or some similar system.

[NullDragon]

Currently the PSR isn't based on just win/loss. If you play a balanced game, you get a standard amount of PSR added/subtracted based on if your team wins or lose. If one side is uneven, then generally the team who should have a harder time gets more points if they win, and loses less points if they lose.

I'm not too much of a fan of this proposed system. Kills really don't have any measure of success whatsoever on your team, from what I can tell. The scout, for example, gets horrendous amounts of kills on new/average people without doing a thing, because of invisibility and fast movement. Should he get a lot of PSR if they never help the team at all, and his teams almost always constantly loses? No.

Probably other examples, but yeah. While yes, your PSR will be brought down if the rest of your team is horrible, you probably will be playing with a wide variety of teams and thus it's still a somewhat good indication of your skill. It's not perfect, but what in life is?

[KGB_Killer]

Ok i read and diggested most of the arguments against the system and i would like to point out if it was unclear that the system is almost evenly balanced between support and carry players. Carries gain extra points when it comes to kills and creep kills, but lack assist. Support players rack up tons of assist and because of this the system pretty much works itself out.

The slight flaw is that if the game goes 70+ minutes the number of points the carry recieves is going to outway the support player because the carry's creeps will be around 4x that of the support as opposed to 35 min games where it is just 2-2.5 times.

So when it comes to 95% of the games the system works itself out.


Next there is the argument about heroes like scout who dont die very often and rack up tons of kills. The only problem with scout is that he gets next to no creep kills, and very little assist. He would actually be one of the worst heroes for this system because he gets all his points from kills; where as other "carries" (even though i dont think scout fits that role, he doesnt really fit any role) rack up tons of creeps also,

this would end up stoping people from pick useless heroes in team fights because of there lack of assist points.

Well i think i covered everything, but i do understand it is hard to get an idea based off of just raw data like that but as i said before i can email you the spread sheet which includes the formula, its long and may take about 15 minutes to decode (uses cell numbers) but if you have used excell before you should be able to get the idea of it.

http://hon.slawed.net/sig2.php?nick=KGB_Killer&img=DementedShaman&style=t&bg=000000&bd=DC143C&tx=FFFF00&va=FFFFFF&tx2=3ABCE5

[KGB_Killer]

I would also like to thank everyone that has voted so far and especially those who have posted, please continue to keep this thread going as if you really want a new system implimented its going to have to get upwards of 200 post to gain attention.


http://hon.slawed.net/sig2.php?nick=KGB_Killer&img=DementedShaman&style=t&bg=000000&bd=DC143C&tx=FFFF00&va=FFFFFF&tx2=3ABCE5

[Slayn]

you also need to consider that the longer the game takes the lesser points you get for example for creep denies
because at the beginning it's harder to deny and harder to kill creeps than in endgames
example:
min 1 - 15 : denies = 1 point
min 15 - 25: denies = 0.8 point
min 25 - 35: denies = 0.5 point
or so

and that transfer to the other stats

cause the longer the game takes the easier it is to catch certain stats like creepkills

[darkeagle1]

I think you have good idea in mind, but as in context it does not convince me enough. PSR could (maybe should) reflect more players skills.

It would be really nice if you could write your formula as pseudo code (or something else that can be posted here) OR host it in web (you also can zip and attach it) and I (or someone else) can do that. At least for me plain numbers don't tell anything.

Ideas that game in my mind:
*) Characters could be grouped in categories. PSR change would be calculated a bit differently for categories.
Example: Scout would get additional points from working as scout and subtracting points from just killing nearly killed enemies (or something like that)

*) Players effectiveness would be calculated from current PSR and players level (players gain experience and levels (If I now recall correctly)). That would be then used for winning / losing change. (Also shown instead of current PSR)

As how to really calculate ones skill I don't know. I don't have long game history from dota/hon, so I'm not so in strategies.

[Slayn]

I think you have good idea in mind, but as in context it does not convince me enough. PSR could (maybe should) reflect more players skills.

It would be really nice if you could write your formula as pseudo code (or something else that can be posted here) OR host it in web (you also can zip and attach it) and I (or someone else) can do that. At least for me plain numbers don't tell anything.

Ideas that game in my mind:
*) Characters could be grouped in categories. PSR change would be calculated a bit differently for categories.
Example: Scout would get additional points from working as scout and subtracting points from just killing nearly killed enemies (or something like that)

*) Players effectiveness would be calculated from current PSR and players level (players gain experience and levels (If I now recall correctly)). That would be then used for winning / losing change. (Also shown instead of current PSR)

As how to really calculate ones skill I don't know. I don't have long game history from dota/hon, so I'm not so in strategies.

hmmm good point!
it somehow have to depent on the hero you picked!
interesting way

[KGB_Killer]

The idea of having creeps worth less later would obviously be an addition to my system, and would improve the accuracy but obviously with my limited resources it would be impossible for me to impliment such an idea (although the creators if possible should consider it).


Keep the suggestions coming :)

http://hon.slawed.net/sig2.php?nick=KGB_Killer&img=DementedShaman&style=t&bg=000000&bd=DC143C&tx=FFFF00&va=FFFFFF&tx2=3ABCE5

[KGB_Killer]

I sent the formula to one person and he made it somewhat understandable, he may be posting it on the thread im not sure.

[Devilhate]

Really wanting to show a player's true skill is kinda far off, all you can hope on is to map what kind of games he play's, which style he play's.
Also as my skill lvl varies from game 2 game, and from hero to hero if i have good PSR is decided how scared i am off taking chances, i've met many people who took pride in winning over me while i took some experimental heroes. if i did this all the time i'd have really low PSR, if i do farm with the same hero that i know how 2 play, over and over i'd be the imba PSR king which pwned noobs all day

[TruMentality]

i vote yes

[Isin]

No. The game is about winning, and that's it. The rest is just unnecessary details. I can go 0/10 with 0 creep denies, and as long as my team wins, me > you.

[Timeetyo]

As was said before, there is but only 1 objective to this game...to win. All of the other stats *MAY* be indicators of this, but in the end they can all be gamed much easier than basing it purely on the one and ONLY objective of the game. There is no way to make a system that is fair to all of the different playstyles out there that are both good and bad. All you will do by adding other variables is make it so people are too busy farming other objectives and no longer care about winning the game. So, to address some issues I see brought up a lot:

1. PSR is NOT the ladder ranking system. It is a tool to give you an idea if you are about to get pub stomped. This means that getting a 100% accurate 'perfect' system isn't that important. Save those complaints for when they introduce their ranked ladder games.

2. PSR takes more than 3 games to sort out. At first you see a large variation in rating as everyone gets close to thier true rating. Give it 20+ games...and even more come retail...and it will do its job. This is really important with the argument of 'OMG I was on this bad team...I was punished because of some other tard'. This is currently a problem because that 'bad player' hasn't had his rating truly fleshed out. Give it a bit of time and you'd have *known* beforehand how bad your team was.

3. With that said, the current system isn't perfect, even after it gets sorted out with some more games. I think the biggest thing that needs tweaked is the risk/reward ratio for playing imbalanced games. I don't have a perfect solution to it, but it definately seems off....especially if there is 1 guy who is much higher than the other 9.

4. Another issue we run into is there just aren't enough players to truly fill a balanced game in a reasonable timeframe. I'd love to host a game that was 1700-1750, for example, but I know it'd take 4 hrs to fill. Give us a few thousand more players and that problem goes away.

[KGB_Killer]

So a metaphore to argue your point that all that matters is winning.

Take the Los angeles Lakers for example, 5 people on the floor at all times similar to HoN, first lets look at Kobe Bryant who is regarded as by far there best player, led the team in every statistical category, carried the team to the championship.

Now lets look at someone such as Trevor Ariza (yeah i havent heard of him either but he has played every game for the lakers this year), played a minor role in helping the team win, and just like Kobe recieved a championship ring.

Now here is what you are saying... These two players on the same team deserve equal compensation for thier efforts (in HoN the reward is PSR, in basketball the reward is a lucrative contract). Now I dont see the lakers jumping towards their check book to play Trevor Ariza 20 million dollars a year and yet you think he deserves the same PSR.

Just because the ultimate goal is winning the game, doesnt mean everyone deserves the same reward, it should be based off of thier contribution.

[Primal_Zed]

Well, he did get a championship ring. Isn't that supposed to indicate that he is a 'champion'? Kind of like what the PSR is supposed to do? (comparing the PSR to a contract is a fail analogy, but then you can't expect to win an argument by analogy anyway)

[KGB_Killer]

Its not a fail analagy because a persons contract is a representation of thier skill level (in most cases), similar to what PSR is supposed to measure; skill level.

As for the champion thing, with that being said would you rather have lebron james on your team or Trevor Ariza.

Sure Trevor Ariza is a "champion" but that doesnt mean he is a better player, it just means he lucked out and ended up on a team that was stacked.

[KGB_Killer]

And BTW winning percentage still accounts for rougly 60% of the PSR in this system, depending on the player it is anywhere from 50-75% of the persons PSR.

[Primal_Zed]

One problem with your analogy is you're assuming the same group of people keep playing together, which obviously is not what the PSR is for. It's for people who go around into mostly random groups in different games. After an adequate number of such games, Kobe's PSR will most likely be above Trevor's, because his skill allowed him to bring more games to victory.

[KGB_Killer]

That is true, but i know tons of people who pubstomp with teams of 5 where 2-3 of the people are extremly skilled players and carry the team to victory every game almost and the other 2 people on the team put together average stats but rack up the same number of victories.

Of course the old system works somewhat for someone who moves around, but my system works for the people that move around, and the people that stay on the same team.

[inew]

How about taking out the creep kill/deny stats :).

[Spector]

PROBLEM (current system):

Winning or losing as a team does not reflect upon the skill of the individual always.

If your team wins and you get 0/10 and 0 denies then you are definitely REALLY bad and your rank should probably actually drop in my opinion - moreso than if you get 0/10 and 0 denies on a losing side. I play with friends and sometimes you will win a game but your mate will do absolutely nothing all game and actually have bad stats in comparison to the rest of the team. Yet if he plays lots of games with us and gets carried to victory does this make him a good player? No it does not. He will still get the same PSR as us though...? This is a flawed system to me, however I don't really suggest a solution because it is a very difficult problem.

PROBLEM (stat based solution):

The problem is, there are so many variables you have to define in order to come up with an individual ranking in such a team oriented game. Even if you do, the weighting of these variables in relation to your PSR would be difficult to determine.

For example, let's say you have categorised heroes into defined roles (carry, semi-carry, support, tank, lane pusher, etc.) how do you define points/"good games" across the roles? It is extremely difficult to say that a 2:2:25 (K: D: A) support with 50/20 (CK:CD) in a 45 min game is equivalent to a 20:7:10 carry with 120/15 in a 60 min game in terms of worth to the team. You just cannot measure these things because they are all relative to: the heroes picked, your team mates heroes, the items chosen by all the heroes, how much damage you did to the enemy/creeps vs number of hero/creep kills, skill of opponents that you are killing/laneing against. A dude that plays thunderbringer may not be participating in ganks much and just using his ulti to ks when his team are doing the hard work. He should actually go down in points for this as it's making it harder for your team to win (by denying the carry of gold + items) but with a system that measures K: D: A your rank may go up? Using the alternate system of determing based on W/L is also flawed in this case because is the guy actually contributing much to the team?

SOLUTION (is there one?):

So what's the solution??? I don't think it's really a simple answer. To get a true reflection of skill you need to do some extremely detailed analysis as DotA/HoN is such a complex game. Some preliminary things that I can think of are:
- Measuring creep kills vs damage dealt to creeps vs hero played vs other players playing that hero (so where you sit on a bell curve relative to others and obviously people who do alot of damage to creeps in the early/mid for their creep kills don't know how to last hit very well .'. probably noobs).
- Similarly for hero kills: vs damage dealt to vs hero played vs rank of heroes killed vs hero killed (again no real reward for people who do nothing but use some spell to get the final hit).
- Gold lost vs deaths (would weed out the noobish players from the non-noob players at the very least; good players don't walk around with a tonne of gold all the time).
- Disables on a hero resulting in death. Ie: disabled hero within X seconds of dying vs hero vs rank.

These seem like awfully specific things and some don't apply to all situations: for example, you DO want your Thunderbringer to use his ultimate when Night Hound has run away invisible on 200 hp even if he is nowhere near the gank. The point of these suggestions is really just to highlight the level of analysis I personally think would be needed to come up with a true reflection of a player's skill and worth to a team.

Something really useful I think would be indicating past DotA experience (as the skills are pretty much directly transferrable) and number of games played. I am sure people have talked about this a lot already but it seems this would be a helpful addition. Eg: Played DotA for 2 years + 80 games of HoN is more than likely going to be a FAR FAR better player than someone that has never played DotA + 80 games of HoN.

Difficult problem! Sorry for wall of text.

P.S. Please do not flame for what I place emphasis on. I have played DotA at a moderate level and have a fair understanding of the game, however, these are just MY perception of what makes good/bad players.

[KGB_Killer]

I like that post Spector, and i agree with you 100% obviously no system can trully measure a persons skill but my argument isnt that my system is perfect, it is just that it is a massive improvement on the current system; which according to your post is correct.

I have no problem with a more accurate system being implimented but this was as good an idicator i could come up with using the information provided in the stats window; im sure the game creators could also tweak my system a little bit to make it work even better (using criteria that i am unable to attain).

Bottom line the current system is not an accurate representation; my system i would say is accurate with a 5-10 percent margin of error for a players skill level.

Using further game criteria they could drop the margin of error to roughly 3-4% (this being intangibles which are impossible to measure.

Btw please dont comment saying the intangibles account for more then 4% of the game cause i know they do but some of the intangibles get translated into tangible results in some form. If I have completly lost you then you should probably just vote yes for the idea i have presented. :)

[Andrioid]

The game is about winning, yes. But if the team isn't balanced in the first place; the ranking makes no sense - and does not reflect the player's skill.

The public ranking should reflect your ability to play in a team, as well as your winning percentage.

I can't really be sure if your formula makes sense or not; but it's a step in the right direction.

The biggest problem with ELO style rating is the team element. Just imagine how the ELO would be for chess if all the grandmasters would create a public game under anonymous nicknames. That would really lower everybody's rating and destroy the point of the system.

In my opinion, the most important part of a team game like this is to stay alive. So I want the rating to punish 'feeders' and award survivors and more importantly people with high assists (and of course kills). Then divide this with the winning percentage and you'll have something workable.

The current system will keep matching 'average players' who get unlucky with teams with even worse teams and by that reducing their chance of winning and by that match them up with even worse teams and so on. In short, the system awards 'public stomping'... or at least until the entire match finding process is based on PSR.

But regardless of what final rating system is chosen. It's important that it's only used for the 'quick-game' match-making. Creating a rating based on public games with random teams is just silly.

[darkeagle1]

The formula of this thread... (I comment it later...)

Variable names:

win_per // winning percentage (from 0 to 1)
los_per // losing percentage (from 0 to 1)
awc // Average Winning Change in Game (for team)
kills // Kills per game
deaths // Deaths per game
assists // Assists per game
creep_kills // Creep Kills per game
creep_denies // Creep Denies per game
length // average game length (minutes)


kills_per_minute = kills / length
deaths_per_minute = deaths / length
assists_per_minute = assist / length
creep_kills_per_minute = creep_kills / length
creep_denies_per_minute = creep_denies / length

Calculations

good_ratio = win_per*30 * (10 + kills_per_minute*10 - deaths_per_minute*10 + assists_per_minute*20 + creep_kills_per_minute*4 + creep_denies_per_minute*6)
bad_ratio = los_per*30 * (35 - kills_per_minute*10 + deaths_per_minute*10 - assists_per_minute*20 - creep_kills_per_minute*4 - creep_denies_per_minute*6)

kills_to_deaths_ratio = kills*(70 - length) - deaths*(70 - length)
awc_ratio = 1400 * (1 - awc)

Result:

PSR = (2300 + good_ratio - bad_ratio + kills_to_deaths_ratio + awc_ratio) / 2

Original excel formula for reference

PSR=(((B2*30) * ((10)+(E2/J2*10)-(F2/J2*10)+(G2/J2*20)+((H2/J2)*4)+((I2/J2)*6))) - ((C2*30) * ((35)-(E2/J2*10)+(F2/J2*10)-(G2/J2*20)-((H2/J2)*4)-((I2/J2)*6)))+((E2*(70-J2))-(F2*(70-J2))) + 2300 + (1400*(1-D2)))/2
where: b2 = win %, c2 = los %, d2 = awc, e2 = kill, f2 = death, g2 = assist, h2 = creep kill, i2 = creep deny, j2 = length

ps. tell if something could be more readable... tryed my best...

[darkeagle1]

Your formula have goods and bads...


Variable names:
These in real implementation should be per game (not all time). I understand that you worked what you got.


good_ratio = win_per*30 ...
bad_ratio = los_per*30 ...
Are all these numbers "random" (make to create good result)?


kills_to_deaths_ratio = kills*(70 - length) - deaths*(70 - length)
As good formula, this doesn't work... It can't be time "limited"


awc_ratio = 1400 * (1 - awc)
This I didn't understand... How can awc be used to calculate psr if awc is calculated from psr?

---
As I thought, formula isn't good enough, but it has good direction.

I would prefer real time calculation in game. Could even effect hero's experience gains. In some heuristic way game would understand your role and what are you supposed to do and reward or hit you accordingly (eq. experience gain or lose), but still keep multiple and unknown game styles possible (I don't want it to limit your creativity).

Game is about to win, but still some player skill indicator is great so random matches with auto balance would create good games!

I leave this REALLY hard and problematic part to s2games. Maybe they have time to sit down in front of whiteboard and think, a lot. They also could use experienced DOTA players for help (at least for game styles and how those are seen from character actions).

If this interesting logic is implemented, it also could be used for tutorials or new player helper. Like telling hints to new players, have more info. etc.

Game could tell player how special role heroes are supposed to be played (So they wont run into middle of fight whit steal character) and maybe disable of using or warn low level players about those. (Don't forget that player own experience and leveling. It can be used also!)

[EclyptO]

This is a much better system, i may be noob myself, but i reckon this'd be much better cause u can be pro, but have a noob team, therefore have a **** score for that round.

[Drunken_Jedi]

Ok i read and diggested most of the arguments against the system and i would like to point out if it was unclear that the system is almost evenly balanced between support and carry players. Carries gain extra points when it comes to kills and creep kills, but lack assist. Support players rack up tons of assist and because of this the system pretty much works itself out.
Sorry, but that's just bullshit. Why would a support hero necessarily have a ton of assists?
Take Glacius as an example.
He won't get many creep kills and denies since he won't spend much time farming. Typically he'll die first in a team fight, so he'll get almost no creepkills and few assists, but lots of deaths. Unless you make a system that somehow accounts for the extra spells that your allies can cast due to Glacius's aura, Glacius will always have horrible stats.

Contrast that with a semi carry like Defiler. She'll get a lot of creepkills, a good amount of denies, lots of kills and assists and typically not that many deaths.

[zzSleeper]

How exactly do you plan on calculating AWC based on those numbers? The accurate calculation of AWC is the sole purpose of a good ranking system and you don't even include how you propose to calculate it. Until we have that your proposal is meaningless, because all you include are some sample numbers with data that fits your equation.

[Rippsy]

We haven't had the current PSR system long enough to see if it will even work yet.

Ranked-Matching is not in yet either; so your really making a lot of fuss over something which might not even need changing.

[zzSleeper]

We haven't had the current PSR system long enough to see if it will even work yet.

Ranked-Matching is not in yet either; so your really making a lot of fuss over something which might not even need changing.
Fully agreed.

[Decency]

This will further encourage players to killsteal, statwhore, and pubstomp. Basically everything that you don't want in a game seeking to have a good competitive environment.

ELO systems (especially ones like this PSR which gives balanced winnings/losings) by their nature take time to work. The system has been in place for something like a week, and it's gotten huge complaints already, which is just ridiculous. PSR takes time to fluctuate to a true rating, but it is a tried and true system and it will work, even in a team-based game.

[KGB_Killer]

I like the chit chat going on here and i would like to address some things.

First of all pub stomping and how it would be stoped under this system, if you were to constantly pub stomp then your teams PSR would be higher by far then a random pub team, therefore making your AWC become much higher. (if you pub stomp usually your about a 75% chance to win and that would result in less points for the game)\

Secondly about the 1400* (1-AWC) equation, basically what this means is that for each percentage in either direction your AWC causes a 7 point drop in PSR ( someone who has an Average Winning Chance of 75% because they pubstomp would ultimately lose 25x7 = 175 points in there PSR. Similarly someone who joins the bad team every game and has a 30% chance to win would gain 140 points. I felt this was a resonable amount per percentage point but of course could be raised or lowered as to whether people see it fit.

Thats all I have to say for now, keep up the posts. :)

[edxs]

Elo is so far superior to a linear weights system that I wouldn't switch to this despite all the current system's faults.

[Helgeran1]

It seems to me that you gain more than you lose on average so in the end you get higher rank just by playing more games even if you win <50% of the time.

[Primal_Zed]

If it's either teams pub stomping or teammates stat padding, i'd take teams going around pub stomping. No matter how much you try to adjust which stats give what kind of bonuses to different heroes, under your system there will always be people who play to stat pad instead of playing to help their team win.

Besides, if you see the other team has a significantly higher average PSR (once everyone has enough games to make the PSR a decent indication of performance), then you don't have to stick around. Whether the other team is made up of people that join random games and happen to be good enough to deserve that stat, or if they are on a team that cooperatively gets that stat, the end result is the same - chances are, they will beat you.

[LegoPirate]

why should game length be included in your psr?

also, i cant tell if longer games have a positive or negative effect on the psr.

you should really make the data more clear. make it so that its apparent what has an effect on psr and how great of an effect does it have, because atm your chart is all over the place and just shows that w/l and k/d are the main contributors to the score.

[BobSacamano]

I'd like to see mine

[eisenpony]

I'm sorry I don't really have anything constructive to say about your system, this is more a general comment on analyzing player skill.

When I am judging a players skill it usually has nothing to do with player / creep kills but how they react in certain situations. A skilled player knows when an ambush is coming, and they know how to move together with their teamates. An even more skilled player is giving advice and calling out plays. Subtle things like crowd control, harassment, and scouting are important to end game victory, but impossible to evaluate with stats.

Ultimately, I don't think you're going to find a solution by crunching numbers. You need an ELO system, or a player rating system. At the end of the match players should rank their teamates and opponents.

[guoguo]

The proposed new PSR rating system is flawed because it explicitly puts bias on the recorded stats while ignoring other skill factors which lead to game wins. Basically, you're saying a good K/D/A and creep stats are "skillful" while unrecorded stats, supporting the team with wards, having good map sense, are not. For example, having good harassment and lane control denies your opponent the opportunity to safely farm creeps. These skills show up in your opponent's creep stats and not yours. Helping an ally escape a gank would do the same. The problem is that it's difficult to program an algorithm which would record "harassment" or "map sense" and even more difficult to score the effectiveness of these skills.

In addition, the weights and scoring of your system I feel are completely arbitrary. What does having 50 more rating points than your opponent mean? The Elo system would tell you that if the two of you were hypothetically matched up a thousand times, you would win a certain percentage of those fights. I recognize that the Elo system is imperfect because it's a reflection of the performance of every team a player has been on rather than his performance in every team you've been on, but I don't think this is the best solution to that problem.

[zzSleeper]

Some of you may not have seen this: PSR and Why It works (http://forums.heroesofnewerth.com/showthread.php?t=8105)

[LegoPirate]

tangibles are often related to intangibles.

sure the things you pointed out arent directly measured, but the results of those intangibles are recorded, such as avoiding ganks means your deaths will be lower, calling out plays and leading a team to victory gives you that W that everyones after. all of those things you both pointed out help the stats that do get recorded.

[HunteR_]

Did you just make a new stat system to take the highest ratings you had fromt he game stats and make a post of it being a valid scoring system ?

The difficulty of the game you play is measured by the PSR of your team and the PSR of the other team. The PSR rating is exactly what is stands for, public skill rating.

Your proposed rating is a public kill steal or public carry rating.

Your method doesn't incorporate any methods of measuring support heroes/players.

Every rating criteria you applied except for assists apply to carry heroes only. Gankers don't creep, gankers don't deny, support heroes baby sit and do not last hit creep to feed carry crucial early game gold. I mean a plain example of a statistic you didn't even think of is to include points for wards bought.

I want a rating system that rewards teamwork, ganking, communication, map vision, helping the team, ganking, and supporting; not kills.

Course I guess that's why people play in league inhouse games, so that every game they play is about picking a carry hero and going godlike.

Someone's real skill level will be determined from higher level ranking systems, pub skill rating, is and always will be a joke, because it favors crushing kids, and not playing as a team/support. Or of course getting your friends on a team and crushing your opponents to pad stats.

Time and time again i see people with ridiculous stats, and no skills. You proposed method, only supports that ideology more.

[eisenpony]

tangibles are often related to intangibles.

sure the things you pointed out arent directly measured, but the results of those intangibles are recorded, such as avoiding ganks means your deaths will be lower, calling out plays and leading a team to victory gives you that W that everyones after. all of those things you both pointed out help the stats that do get recorded.

They help, but they don't define. A great player could call out ganks and defensive actions but take part in very few of the plays himself. How does helping a teamate escape a death reflect on my stats? How about telling someone to back off because a gank is headed their way? Letting my agi teamate grab the DD rune while I run in as bait? Only the player I helped gets a bonus. A great player could sacrifice himself to let his teamate(s) live in certian tactical situations. These are things you aren't going to be able to calculate.

[LegoPirate]

you get the W, which is still one of the largest factors in kg's psr

[Spector]

PSR completely broken because of the weighting so heavily on wins and losses.... the idea is just flawed. You can have the world's best player in the world's worst team and he will have the poorest PSR in the world if he continues to play with them.

My PSR dropped from 1650 -> 1470 odd from 4 or 5 straight losses last night because of being on teams of players that are mega n00b or the host being a knob and stacking the teams. (Had one guy join a game "AP Pros" and was 0 9 after about 10 mins, it was his FIRST EVER game of HoN and when I asked "Why did you join a pros game?" he replied - "I didn't know if I was pro or not".)

The amount of poor quality games is becoming ridiculous. I think I have had maybe 1-2 decent games since PSR introduced...

[Decency]

PSR completely broken because of the weighting so heavily on wins and losses.... the idea is just flawed. You can have the world's best player in the world's worst team and he will have the poorest PSR in the world if he continues to play with them.


No, he won't, for two completely different reasons. You should probably stop posting if you don't understand even the basics of the system (that goes for at least half of the posters in this thread).


- First of all, a player with a PSR that is far outside the average, such as "the best player in the world" would be risking +0/-0 because his PSR would be too far distanced from the average of the rest of the game's players.

- Second of all, his team will have a poor average PSR, meaning their chance to win will be lower. Because they will have a less % chance to win, they will have a likewise balanced amount of points risked, for example: +37/-3.

[uberdabing]

Read my sig please for a very detailed explanation of why the current PSR is the best system available.
http://forums.heroesofnewerth.com/showthread.php?t=8105

[lcite]

Def want a different system then the current PSR. Dono if yours is the one to do it, but incorperating assists, creep kills/denies etc is a step in the right direction.

[Spector]

No, he won't, for two completely different reasons. You should probably stop posting if you don't understand even the basics of the system (that goes for at least half of the posters in this thread).


- First of all, a player with a PSR that is far outside the average, such as "the best player in the world" would be risking +0/-0 because his PSR would be too far distanced from the average of the rest of the game's players.

- Second of all, his team will have a poor average PSR, meaning their chance to win will be lower. Because they will have a less % chance to win, they will have a likewise balanced amount of points risked, for example: +37/-3.

Ok let me rephrase. You have a person that has played DotA for 4 years and has a VERY GOOD understanding of the game... he is not playing with mates he is playing with randoms and currently has a PSR of 1500. He gets stuck with crappy team mates who have never played DotA and are in their first few games of HoN also at equal PSR. He keeps getting teamed with these sorts... PSR becomes something abysmal and has no reflection of the player's actual skill.

The system works if you have a bunch of mates who are good at the game/you get paired with randoms that know how to play. It does not work for people that are good at the game but get stuck on crappy teams.

In fact, because of this, people that are actually decent at the game are getting kicked from "Pro" games because their PSR is too low. Also, there seems to be alot of **** hosts out there that try to stack teams such that the good players are all one team... so you end up sometimes with "good" players getting stuck on **** teams all the time and constant crappy games.

I do not claim to be the best at this game and neither do my friends but we have ALOT of experience from playing DotA over several years and honestly, since the PSR system has come in there has been nothing but ****, uneven games.

I am sure there are many other people in this circumstance and this is an example of the system not working and consequently making people not want to play the game...

[Spector]

Also, I have read your thread on why PSR works and disagree with several of your points because they make certain assumptions about the people in them (namely your first three "complaint"/"answer" points).

What I am complaining about is the attitude coming along as a result of the sytem:

"As long as you don’t voluntarily participate in pub stomps or horribly unbalanced games, this is the inevitable law of large numbers."
It is becoming increasingly difficult to find these games since PSR has been introduced. Lots of people just want to win.

"This means that even if your team is not amazing, if you are truly good, you can make up for bad play by your team. Of course, this does not mean you can carry and 1v5 the enemy, but it means you will lose LESS games on average due to having bad players."
I think this shows a lack of understanding of the game. Take this in consideration with the above and below points.

"I have never heard of people getting kicked for being too good, unless it was for a noob only game."
Happened several times to me in a short space of time yesterday and been happening every day. I think I left 4 games in a row on countdown because host kicking higher ranked and then not balancing.

I am not trying to flame here or anything. I am merely providing counter arguements from real cases as to why I believe the system does not work that well.

[Decency]

Ok let me rephrase. You have a person that has played DotA for 4 years and has a VERY GOOD understanding of the game... he is not playing with mates he is playing with randoms and currently has a PSR of 1500. He gets stuck with crappy team mates who have never played DotA and are in their first few games of HoN also at equal PSR. He keeps getting teamed with these sorts... PSR becomes something abysmal and has no reflection of the player's actual skill.

The system works if you have a bunch of mates who are good at the game/you get paired with randoms that know how to play. It does not work for people that are good at the game but get stuck on crappy teams.


Yes, if a very good player just happens to get stuck with complete newbs for his first x games, his PSR will drop. Then, continuing to play at the same level of games, he will begin to win when he stops playing with complete newbs at x+1 games, and his PSR will rebound more dramatically because of his low rating.

Please Google Law of Large Numbers. If you're going to use such a ridiculously rare occurrence to try to prove a point about the validity of a system, you're better off not posting.

No ranking system is ever going to be perfect, but the PSR is one of the best I've seen.

[Spector]

What I am trying to explain here is basically what I think VortexMagus is explaining in the other thread uberdabing.

[Spector]

Please Google Law of Large Numbers. If you're going to use such a ridiculously rare occurrence to try to prove a point about the validity of a system, you're better off not posting.

I am not a retard I have a high level understanding of mathematics, please don't patronize. Also, please read with relation to the rest of my post. The problem is the possibility of this happening coupled with the attitude of the community at present which is causing either: longer search time for games (having to leave because of imbalance of teams in order to find an evenly matched game) OR having crappy quality games (because you have settled with a pub stomp/bad balance in teams just so you can actually have a game -> PSR dropping).

[Andrioid]

Yes, if a very good player just happens to get stuck with complete newbs for his first x games, his PSR will drop. Then, continuing to play at the same level of games, he will begin to win when he stops playing with complete newbs at x+1 games, and his PSR will rebound more dramatically because of his low rating.

Please Google Law of Large Numbers. If you're going to use such a ridiculously rare occurrence to try to prove a point about the validity of a system, you're better off not posting.

No ranking system is ever going to be perfect, but the PSR is one of the best I've seen.

What about an 'average' player that gets stuck in newbie games? Then it would take him way longer to 'bounce back', correct?

In theory if all these matches were being auto-balanced; then the PSR will in the end become more accurate.

However, not all matches are. If you look at my stats, I'll stand by my KDA but my win/loss just keeps plummeting. I am not a 'pro' player but I'm not new to HoN or DotA either.

IMHO, this isn't necessary a problem with the rating system. Maybe if feeding of one player on the team couldn't tip the balance of the entire game, this would work perfectly.

[Decency]

What about an 'average' player that gets stuck in newbie games? Then it would take him way longer to 'bounce back', correct?

For every game he plays, there's an equal chance of him being on a newb team and of the other team being a newb team. Except for if the player contributes to how good/bad his team is: however much he contributes leads towards finding an accurate rating, known as PSR.

What Kills:Deaths ratio exactly are you standing by? You seem to be a weaker than average player, by your stats.

http://hon.slawed.net/sig.php?nick=andrioid&img=Magebane&style=t&bg=000000&bd=00FF66&tx=009900&va=CBD3F7&tx2=0066CC

I would assume from your Wins/Losses that your PSR reflects this, as well.

[uberdabing]

Problem is not PSR. Problem is no easy way to find players of equal skill, aka no matchmaking. Be patient and wait for the patch.

[Heliotic]

If your win loss is bad and you are playing in autobalanced games, the common theme to all the games is you. If you'[ve played 50+ games and still have a low PSR, you probably deserve a low PSR.

[Diplococcus]

Why even have a PSR, public games are just public games and in the long run don't matter. All this PSR system has done is create arguements and unneccesary whining. It was better before when nobody knew how well other people were outside of their friends so they actually randomed teamates instead of basing it off of some skewed point system. When it comes to playing with friends, I don't think you care what your friends PSR ratings are anyways, and I doubt you care what the other teams chances of winning are. PSR gives people reason to complain in pure public games, and doesn't matter when it comes to just playing a few games with friends. Save the PSR for ranked matches, and let the public games be casual the way they were meant to be.

[Diplococcus]

Casual public games are casual, lets leave them that way. Its annoying enough when you lose a game, now you have to look at a shitty score to remind you of your bad luck? That's not very fun/casual to me. PSR = too serious and drains the fun away from public/casual matches.

[HellbirD]

Please make sure you don't chock it up just to K/D

There are other very important stats, Damage done to enemy heroes, damage done to towers, damage done to creeps, consumables used, wards placed, abilities used.

Perhaps people can suggest some more, the more statistics we can incorperate the more accurate the system will be and most importantly, would in some way prevent issues arising around killstealing etc, because people would want to use their spells to do as much damage as possible etc.

[RaWRpewpew]

I personally dont think there is much wrong with the system, however *IF* there is to be a change i would say it would be something like the following.

Using PSR before game you calculate prelimimary +x/-y

Then since every player will never equally contribute to the game, a range of +/- gets calculated so instead of +20/-10 it gets changed to
+(15-25) / -(7.5-12.5) [Using a 25% change, I actually think a smaller % should be used here]

Then using a combination of your stats in the game compared to the average plotted on some kind of distribution graph.

I think the way to go is keeping PSR like it is but adding a small stats related change to your final PSR adjustment after the game. To soften a loss for a better then average player in the game, and making it worse for a below average player in the game.

It could also be made even simpler, by only rating people as : below average - average - above average in relation to a certain game the process becomes alot easier.

Then you get +10%/+0%/-10% to your PSR adjustment depending on how you get rated.

~RaWR