I can't understand the fluctuations of the new ELO system. Yesterday, after three games won I had 1594 ELO points. Today I haven't played yet. At 10.00 am I open the game and ELO rose to 1601. Tonight at 09.22 p.m. fell to 1591. And I haven't played yet. Is it connected with the swings in the stock exchange according to MIBTEL or with Dow Jones? :)
+7 / -0
|
Yes. Just remember, buy low, sell high!
+4 / -0
|
AFAIK, it can move without you playing games, as its based on other players, i suppose if some player beat you once, and they suddenly lose lots of games, yours may rise etc... There is also the possibility Dein is still fucking around with it :)
+0 / -0
|
The developer team has taken your issue into consideration. We have added live stock exchange graphs so you can decide exactly when to invest and optimize your profits. Try it now!The free version is limited to statistics from the last two years, buy the stone-age DLC to see it all.
+8 / -0
|
This was one of my concerns when people started suggesting the new WHR system. It really shouldn't be called ELO anymore. It's not a good system to keep score. The point of WHR is to predict your skill as accurately as possible. That means that your skill will be constantly updated as time passes and people you played against play other matches. The system will know more about them, and in turn, know more about you. To give an example, imagine you play against a new player called Giuseppe, that no one knows anything about yet. Now the system may give him an average rating, and the first matches you play against him, you may lose a lot of points. However, as time goes on, it might turn out that Giuseppe is actually a pro Starcraft player and very good at ZK. After that data comes in, the system will penalize your previous losses less harshly, because it turns out you were playing against a pro. Basically, this makes WHR very poor at showing a score. You can't play games and get a desired, fixed rating. When you wake up, things might be completely different.
+1 / -0
|
Conduct a Zero-Koin ICO on the Ethereum blockchain as a crypto token based on proof-of-skill.
+3 / -0
|
|
|
uh ... dear Dein Freund, I really liked the wall street chart, but, if you have to, next time, if possible choose less strong players to make a comparison ... :)
+2 / -0
|
  manero you can compare anyone with that tool.
+1 / -0
|
Why does my rating chart only go back to 2015?
+0 / -0
|
WHR is currently only processing battles from the last two years to estimate your rating. I'm currently experimenting with extending that to 5 years, but there are some weird issues cropping up: 
+1 / -0
|
quote:
That means that your skill will be constantly updated as time passes and people you played against play other matches. The system will know more about them, and in turn, know more about you. |
It doesn't know more about my previous games when it gains knowledge about other people's games *now*. Or am I misunderstanding that? Say a person starts to play and gets some initial rating (I didn't read the paper about WHR, so please correct me about how things are!). That rating will have some large uncertainty, of course. I play against her and our ratings are adjusted. The adjustment to my rating should be given a rather large uncertainty, and the new player's rating is also adjusted and her rating uncertainty is lowered. Now let's say I lose against the new player and my rating was decreased a little. The new player now plays a lot of games and her rating increases, and her rating uncertainty decreases. How can this possibly affect my current rating? The new games the player played can't say anything (much?) about her skill state in the game that she played against me in the past. There must clearly be a time component in this algorithm, and I assume there is? Like how long the correlation length is. How do you set these constants to sensible values?
+0 / -0
|
The algorithm expects your rating to behave as a Wiener process when going form one day to the next. I have to set a constant for this process that describes how much your rating is expected to change between any two dates. Thus, a new player is only expected to have changed "so much" since his first games. This can be used to adjust his original rating. I have chosen this constant so that it predicts as many battles accurately as possible.
+1 / -0
|
Random thoughts: - The curve allows you to define a learning efficiency in a certain time interval as derivative of the curve divided by the number of games in that interval. - Figure out an algorithm to determine the players to play with to maximize your learning efficiency => teaching efficiency => profit???
+0 / -0
|
quote:
WHR is currently only processing battles from the last two years to estimate your rating. I'm currently experimenting with extending that to 5 years, but there are some weird issues cropping up:
|
wait what?
+0 / -0
|
|
|
Would you be able to include uncertainty bands in the graphs?
+0 / -0
|
|
|
Godde was remarkable, but that's historic, it's like comparing 1960's sports stars with today's on stats. especially in casual\team rankings a svatopluk or sigero joins that league. Sak: yes, why aren't u ranked yet, i choose not to play team games with you again until your rank kicks into gear :)
+0 / -0
|
I also question the WHR accuracy in terms of current value, some players shoot through the roof initially, and some play mediocre for years but suddenly find a gear or 3. With this ranking would the quite comparative player with a history of average results simply score less because their history weighs them down?; despite an equal current competitiveness?
+1 / -0
|
