|
On June 17 2012 21:24 korona wrote:Show nested quote +On June 17 2012 21:06 enemy2010 wrote:So for me there are the following graphs: CLICK I am platinum myself. So when my graph falls under the green line (which is platinum), does that mean i get demoted to gold? Or to wich value has my graph to go until I get demoted? Same for promotion: Do I get promoted to diamond when my graph crosses the next (blue, diamond) line? No. The demotion/promotion happens when the moving average of your MMR stabilizes itself inside some skill tier in the lower/higher league. More info: http://www.teamliquid.net/forum/viewmessage.php?topic_id=195273It surely would be interesting to know how many games Blizzard includes in the moving average...
On June 17 2012 21:24 Alexj wrote:Show nested quote +On June 17 2012 21:06 enemy2010 wrote:
So when my graph falls under the green line (which is platinum), does that mean i get demoted to gold? Or to wich value has my graph to go until I get demoted?
Same for promotion: Do I get promoted to diamond when my graph crosses the next (blue, diamond) line? It's not cut and dry. It just gives you an estimate that you're close to promotion/demotion
Thank you both very much 
|
On June 17 2012 21:06 enemy2010 wrote:So for me there are the following graphs: CLICK I am platinum myself. So when my graph falls under the green line (which is platinum), does that mean i get demoted to gold? Or to wich value has my graph to go until I get demoted? Same for promotion: Do I get promoted to diamond when my graph crosses the next (blue, diamond) line?
yes. but not instand. You must be stable above or below a line to get demoted / promoted
On June 17 2012 21:09 etherealfall wrote:
It's odd, now that i've played 3 more games, my MMR is back at 1850+.
Does that mean my data is 100% now?
nothing on the world is 100%
its as good as i can tell
On June 17 2012 21:16 TechNoTrance wrote:
Works great! Amazing job man.
Thank you.
On June 17 2012 21:18 Peleus wrote:
Not sure if it's normal for the pretrigger to run on replays? I don't think it matters much either way, just curious if this has always been the case.
Cant detect what you do ingame. I just trigger if you are in a sc2 game. a replay or nexus wars. Because the postrigger dont run nothing happens, he catch a profile print something in the log and done.
On June 17 2012 21:22 Alexj wrote:I am still not sure what is a "good" game? For example: Show nested quote +15:4:37 Webgrabber: Player true
15:4:41 Webgrabber: Opponent true
15:4:41 DATA Alg.B: failed, No Bonus Win
15:4:41 Adjused Calculate Player:true
15:4:41 Adjused Calculate Opponent: true
15:4:41 Gamecheck: true
15:4:41 POST-Trigger done
15:4:41 ERROR: Analyser F: Need more/better games!
15:4:41 Player-data-Qualitiy: 75%
15:4:41 Opponent-data-Qualitiy: 75%
15:4:43 OK Uploader: game uploaded
15:4:43 Ladder Game Done!
15:4:43 -------------------------- I won that game, and I had bonus pool, and I had memory analyzer on
no
still not good enough. Data collection went good, but there is something in the data he dont like.
Its to much deviation or most likely your mmr is caped because you played a guy who is in higher tier than you and your mmr is under the minimum mmr of his tier. All i can tell you than is that your mmr is between 0 and the the top of the league of your opponent.
Not very useful.
Blizzard did this as workaround so none ever drops to 0 points.
bad walkaround if you ask me...
100% is a game he needs to use is as base for all the games under 100%.
|
Well, after playing a few games, new algorithm seem to work great. Thanks again for your work!
|
Hey Skeldark, after the last update the graph without an analyzer is really funky looking:
![[image loading]](http://i.imgur.com/bBM00.png)
It used to look more broken up, and certainly not plummeting like that. This is without a moving average. I know without an analyzer, it's not very useful to me, but I thought I'd point it out.
|
On June 17 2012 23:14 InfCereal wrote:
Hey Skeldark, after the last update the graph without an analyzer is really funky looking:
It used to look more broken up, and certainly not plummeting like that. This is without a moving average. I know without an analyzer, it's not very useful to me, but I thought I'd point it out.
dam you!
why do people have to enable "advanced options".i should have call it "bad options".
The truth is : i know. But after writing the new analyser and fix all the bugs, i notice this one, put it on the todo list and just uploaded the version. you got me ^^
|
United States12235 Posts
I suggest adding colored dashed lines to represent each league's division tiers. This will have several benefits. First, it will allow us to track which tier a player has been promoted into, and future tracking will allow us to prove whether the guessed destination tier was accurate. Second, we can measure the slope of the MMR curve to get an idea as to what the exponential moving average function which governs promotion stability is.
|
On June 18 2012 01:35 Excalibur_Z wrote:
I suggest adding colored dashed lines to represent each league's division tiers. This will have several benefits. First, it will allow us to track which tier a player has been promoted into, and future tracking will allow us to prove whether the guessed destination tier was accurate. Second, we can measure the slope of the MMR curve to get an idea as to what the exponential moving average function which governs promotion stability is.
Didn't we do all of the work to get rid of tiers and leagues?
I wrote this program to show the real skill value and not the fake point system.
I analyse tiers and leagues to calculate them out because they are just errors in a real skill system.
So i dont want to add something like division tiers into the graph for the enduser. I would even remove the league lines because they are nothing else than little images at random skillmarks, but people are sadly used to them. Also promotion and demotions are pointless shadows of the fake skillsystem. What i want to archive with this tool is that people forget their meaningless points and bnet icons and not that they think about division tiers.
The goal is not to tell someone when he gets promoted. The goal is to show him, that its a random mark and having 5 more mmr points then yesterday is the exact same as an promotion.
Just without a little star picture and a "powerful" name like platinum or "diamond" or "master" and way more accurate.
For analyse reasons this is a good idea, but not for the end-user-program. The problem with it is, you use my algo to show the mmr, decide the league of his promotion by this and than decide the error with the same algorithem. So the algo will judge himself always correct.
I think more about makeing a table of all tiers of everyone he judged and look if on different graphs, the algo. judge the same division into different tiers. Also algorthiem A works completely different than F. A crosscheck perhaps shows mistakes.
The promotion / demotion question is because of above reasons not very interesting for me.
|
United States12235 Posts
I disagree, I think it's important to show both. The leagues -- and arguably even the division tiers -- are not meaningless, they still represent tangible goals for players to work toward. The reason for my suggestion was twofold: it allows for error checking so we can confirm the offset data we have is actually correct, and it provides end-users with the information they're looking for anyway, which is "how close am I to a promotion?" Even if we had a perfect model of the entire spectrum, all players want to know is when they'll get promoted because that's something within the game that they're looking for.
|
On June 18 2012 03:17 Excalibur_Z wrote:
I disagree, I think it's important to show both. The leagues -- and arguably even the division tiers -- are not meaningless, they still represent tangible goals for players to work toward. The reason for my suggestion was twofold: it allows for error checking so we can confirm the offset data we have is actually correct, and it provides end-users with the information they're looking for anyway, which is "how close am I to a promotion?" Even if we had a perfect model of the entire spectrum, all players want to know is when they'll get promoted because that's something within the game that they're looking for.
I sort of agree with Excalibur. I want to know my mmr so I can know if I'm improving or not. As I previously noted I promoted to gold this season partly because knowing my mmr had plateau'd in silver motivated me to push myself harder and figure out how to improve my game.
Also, people do need Interim targets and goals. At work, if the bosses just push for the end goals of a large project, people lose motivation. You have to set up interim targets and rewards.
Leagues do provide that and to a lesser extent division tiers could support that if known (think about how any amateur sport divides its players into skill groups).. Also leagues provide a handy way for people to communicate their current ability(in a way even non-players can understand).
I can't believe I just defended the league system, but it does serve a role. My frustration is the point system doesn't provide me with a way to assess real progress.
|
My only complaint with the league system is how it hides MMR. I'd prefer to see Gold 1250 instead of just Gold. This tool allows us to see more details about our skill level, but we'll never totally escape the league system...
|
first explanation for the people who misunderstand tieroffsets: Tieroffsets are not like Sub-leages. You can not get promoted/demoted between tiers. This means your tier hold no value over you current skill"
My analyser dont even know your tier, or try to guess, or care for it to calculate your mmr.
To confirm the offsets looking at graphs is not a good way. We looked at many graphs and calculated many.
"I have the feeling this offset is ok" dont help much. I did this in the past. I now run analysers over the data and crosscheck with different algorithm and let them print objective errormargins. I understand leagues as a goal for someone , like if you say to yourself "i want to reach 2k elo" in other systems. In the end i put the lines in. but why tiers? showing the tiers dont show you when you get promoted. All this data is already extracted and in the graph. Why putting calculation steps in the end result?
.Analysing the data is the way to confirm offsets not looking at user graphs. But i can and i do look at every graph from every user. I can see all graphs with one mousebutton. I do it to search for obvious mistakes in the analysers or to check new analysers for bugs. But to confirm offsets and the tier guess, this is the wrong way. We know already that our offsets are 40 to big between bronze and master. I can prove that not a single offset is wrong by over 30 alone. the offsets we have right now a near perfect but not perfect.
I can tell some single tiers and check if the analyser agree with me on them. I can not tell this by the graph because he includes to less information for that. All you see in this program is the result.
This tool collect data but mostly it shows the result. Its not a real helper for the way. I have different programs for that.
Tools that can run the analyser over the hole datafile or search for some special cases that could give us more information.
Notthat does the same in exel.
If you escape the league system or not is in the end a question you must answer yourself. You can call it 1630 or gold or 54% skill/population. Its all the same, just different accurate.
However i don't understand why the normal sc2 player want to see tier-offsets.They don say anything about the player skill, they dont give any useful information over promotion/demotion.
Short: there is not information that tieroffsets give you, that is not already in better way in the graph. I mean we publish all this data, i let my analyser output every step and every guess. I upload all data that we collected, clean it and public the analyse. Notthat makes graph and long articles about everything he find out. If you are interested in the theories everything is there. the tieroffsets are extra human readable in the bnet.data file you download with the client.
But putting this data in the end result is equal to agree with what blizzard did.
I show the clean skill function that is behind all of this and try to be as accurate as possible.
For this i sadly have to analyse a strange hiding function to calculate it out. For me offsets are just errors in the system.
|
I agree with Excalibur_Z, ttfdrevil and Cheshyr. The leagues serve a purpose that the mmr should do, but naming things instead of showing numbers is what blizzard decided to do. because is easier to remember or compare in a large scale but meaningless between two players that have similar skill.
They both serve two different purposes, the league to classify people in big chunks and the mmr to know the real skill. I think is pointless to not show the tier lines, people (me included) want to know where exactly is my skill compared to others and the only way to know it is using leagues. And know in which tier you are.
|
On June 18 2012 04:58 dareKEU wrote:
I agree with Excalibur_Z, ttfdrevil and Cheshyr. The leagues serve a purpose that the mmr should do, but naming things instead of showing numbers is what blizzard decided to do. because is easier to remember or compare in a large scale but meaningless between two players that have similar skill.
They both serve two different purposes, the league to classify people in big chunks and the mmr to know the real skill. I think is pointless to not show the tier lines, people (me included) want to know where exactly is my skill compared to others and the only way to know it is using leagues. And know in which tier you are.
Like i said, knowing your tier say nothing about your skill. Absolute Nothing!
When i put the tier in the result people will start to compare it. They think that a guy in higher tier have more mmr, but thats not the case. All it say is that at one point in the past this was the case and this point is even different for both.
Also leagues are a very inaccurate system. Its a bigger chunk than you think:
gold = between low silver and high plat. = 60% of the hole population.
If you are in gold you know you are not the down 20% and not the top 20%. Thats all.
Someone in silver can be better than someone in platinum without problems.
|
Technically, MMR is valueless as well, since this is a competitive system. Every game played changes the value of a point. A 5 point gain doesn't actually translate to improvement. Let us have our tiers. The incremental than motivation, and fancy pictures have proven to have other benefits than a pure measurement of skill.
|
On June 18 2012 05:06 Cheshyr wrote:
Technically, MMR is valueless as well, since this is a competitive system. Every game played changes the value of a point. A 5 point gain doesn't actually translate to improvement. Let us have our tiers. The incremental than motivation, and fancy pictures have proven to have other benefits than a pure measurement of skill.
Tier have nothing to do with current skill.
I dont use your tier for calculate your mmr because there is no value of your skill in it.
I use the tier of the opponent! So do you think how good HE played his ranking games have something to do with how good YOU are at sc2?
No it have not. But the system leaks information over the change-points adjusted-points and tiers that i can than use to find the real mmr behind all of this.
I think most of this comes down to a misunderstanding what a division tier is.
It is NOT a subdivision.
|
I think we're all throwing around words and not being explicit. I don't care about tiers. Promotion and demotion across leagues is of interest to me, as well as my calculated mmr. I think these things have value as motivation and broad categorization. That all this system will ever accomplish, which is fine since that's all it needs to do.
All I'm asking is that you leave in the league boundary stuff. Its analytically meaningless, but still part of the game.
|
On June 18 2012 05:25 Cheshyr wrote:
I think we're all throwing around words and not being explicit. I don't care about tiers. Promotion and demotion across leagues is of interest to me, as well as my calculated mmr. I think these things have value as motivation and broad categorization. That all this system will ever accomplish, which is fine since that's all it needs to do.
All I'm asking is that you leave in the league boundary stuff. Its analytically meaningless, but still part of the game.
i have them in, i let them in. Excalibur suggested to add new lines for division tiers. Thats what the discussion was about...
I dont like leagues but i understand that after 2 years people are so used to it that i have to accept them.
|
United States12235 Posts
On June 18 2012 05:33 skeldark wrote:Show nested quote +On June 18 2012 05:25 Cheshyr wrote:
I think we're all throwing around words and not being explicit. I don't care about tiers. Promotion and demotion across leagues is of interest to me, as well as my calculated mmr. I think these things have value as motivation and broad categorization. That all this system will ever accomplish, which is fine since that's all it needs to do.
All I'm asking is that you leave in the league boundary stuff. Its analytically meaningless, but still part of the game. i have them in, i let them in. Excalibur suggested to add new lines for division tiers. Thats what the discussion was about... I dont like leagues but i understand that after 2 years people are so used to it that i have to accept them.
Yeah and I just wanted the tiers in for analytical purposes so we could compare the offsets and see a player get promoted by stabilizing within a tier. If that's not going to be accurate and there are better ways to error-check the offsets then that's fine, the tiers don't have to be shown. I do think it would make explaining the existence of tiers easier though (since you would actually be able to see them on a chart) but it's no problem.
|
Ahh, I misunderstood. I think we're running into a difference in needs between people using as a gaming tool and people using it as an analytical tool. I can definitely see why Excalibur would want tier info, even if that info is novelty for me.
|
On June 18 2012 05:33 skeldark wrote:Show nested quote +On June 18 2012 05:25 Cheshyr wrote:
I think we're all throwing around words and not being explicit. I don't care about tiers. Promotion and demotion across leagues is of interest to me, as well as my calculated mmr. I think these things have value as motivation and broad categorization. That all this system will ever accomplish, which is fine since that's all it needs to do.
All I'm asking is that you leave in the league boundary stuff. Its analytically meaningless, but still part of the game. i have them in, i let them in. Excalibur suggested to add new lines for division tiers. Thats what the discussion was about... I dont like leagues but i understand that after 2 years people are so used to it that i have to accept them.
my point was same as cheshyr. League is a crutch for us mere mortals and a point of reference.
I agree that tiers on the graph are nice to have, but not necessary except to assist excaliber and notthat. I can know my tier (to the limits of the a algorithm) from looking at you data tabs. That is good nuff for me. Hopefully that stays as well.
My point was more philosophical, but I do support excaliber's effort to nail down the division tiers as a challenge and point of interest.
|
|
|
|
|
|
EPT
