Replies: 7 comments 26 replies
-
Total amount of wins will eventually end up in very static line. Downscaling amount of wins to certain value will make map bias more dynamic. Simulate every game as it was x-th game (100th, 200th, 500th etc). Using value of 100 will cause every game around 0,5% change in map bias, 500 will cause change of 0,1%. And it start and continue in loop not losing dynamic effect. Also eliminating big fluctuation in start of data and retaining fluctuation in huge numbers. Example: Current balance is allies: 250 wins , axis 150 wins. Match limit: 400 downscale to 400 Allies wins: 250,3741, Axis wins: 149,6259 1. Downscale (Orange) vs no downscale (Blue) (default) Blue line consists of 5000 events, each event simulating 1 match with 50% win rate. Moving towards static number. Orange line is same data but after 100th match each next match is dowscaled to 100. So each match from that point on behaves as it is 101th match. Those fluctuations are actually one team winning more but due to huge amount of data not making any change in bias (blue vs orange). 2.1. Downscale to 100 Blue is default, starting from 0 wins to either team. In start big fluctuations up to 5% and rapidly going down and eventually end up in no fluctuation. (No more change in bias) 2.2. Zoom in on 100th event Matches with games around 100 as can be seen on graph. Fluctuation value always around 0,5%. And from this point on same characteristics keep repeating. 3.1. Downscale to 500 Orange is downscale to 500, starting from 250wins to allies and 250wins to axis and then downscaled each game. Fluctuation value always around 0,1%. (Always change in bias, but smaller compared to downscale 100). 3.2. Zoom in on 500th event Matches with games around 500 as can be seen on graph. fluctuation value always around 0,1%. And from this point on same characteristics keep repeating. 4.1. Change in win rate, downscale to 500 Here is a data of 5000 events
Orange line is downscale to 500. It can change and reflect true events much more accurately. Compared to huge chunk of data that cant even make it to positive ratio last 2500 games have been +10% chance to win. 4.2.1. Change in win rate, downscale to 100 Downscale to 100, fluctuation is bigger but also adjustment happens much more quickly which can be seen in first matches (1-100). At 250, where win ratio swap happen it can reflect new accurate value in about 100 matches (which is perhaps around month in time period). it does hit +10% quite quickly and keeps fluctuating there as pretty versatile to change. Also fluctuations from start (simulating perhaps new map) are also not that big, rather gradually moving towards real value from balanced state compared to old blue data line which has very big ranges in the start as it starts to fixing around value and increase precision on every continues match Old data chunk cant even make it to positive ratio yet last 2500 games have been +10% chance to win. fixed on 100 managed to swap around in about 100 games while reflecting the 20% change in input yet old data line only raised few % 4.2.2. Change in win rate, downscale to 100, zoom in on 2500th event Such downscaling eliminates huge bias changes in new maps as there is not enough win/lose data and each event causes big change. 8. Change in bias downscale vs no downscale Here is a 25% win graph for one team, zoom in of events 2500-2600. Orange line is downscale to 100 Downscale: Red marked circles show first -0,7% lose then about 10 wins with +0,2% and then 2 loses -0,7% so fluctuating around 2%. No dowscale: old data that is stuck under a 0,1 line so it means over 100 games there is no change in bias. It is mathematically very easily achievable and can apply to old big numbers as well and down or upscale to desired value |
Beta Was this translation helpful? Give feedback.
-
So this is more correct now. Blue line is close to how it is right now. Orange line will take over after 200th event. It is 0,250% change in map bias in a 0,000 balance situation. This also means 4 wins in row will make +1,000% change in balance. When map balance is +25,000% for one team. it is 0,125% when win -0,375% when lose. Lose 4 games in row and map bias will be +23,500% If this feels too much change, value 500 can be used. (0,250 becomes 0,100) Also i think it would be nice to use 0,00 decimal place for map bias too, just like SR does. So it will reflect changes more accurately |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
at 200th game there should be for ecample120wins for one team 80 for other. This is the first previous value in last amount of total wins i was refering to |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
Real live balance state is equal to AXIS[win%] + map bias % if i understood this correctly? so AXIS: 69.3+9.9=79.2%, ALLIES: 30.7-9.9=20,8% or those numbers are in different scale and can not be compared? for example if allies have map bias +20% and real time win rate estimate is 30%. So does that mean this is balanced state for current moment? 20+30=50% |
Beta Was this translation helpful? Give feedback.
-
Using total amount of maps played and wins in map bias calculation will eventually end up in static number once total number goes big enough.
I think using last played maps for the data is more accurate as it will display recent state of map in the server. It will update if one of the teams would start performing better for whatever reason.
For example gathering info only from past 100 maps gives around 0,5% fluctuation after each map and make the number more dynamic and reflect current state of map better.
For example 1000 maps fives fluctuation about 0,05%. Once the number goes big enough it becomes very static and will not show current state of map but rather static number gathered over long period of time.
This graphic below will illustrate how the number will become linear and harder to change
Beta Was this translation helpful? Give feedback.
All reactions