Ballistic Pro Ladder
- iRobot
- Junk Administrator
- Posts: 3909
- Joined: Fri 06 Jan , 2012 10:37 am
- Contact:
Re: Ballistic Pro Ladder
Skill is now just a flat value based on the weapons you use. If you use MARS-3 for example, you're skill 3.5
-
- Disappeared Administrator
- Posts: 4196
- Joined: Fri 19 Mar , 2010 1:21 am
- Location: Earth
- Contact:
Re: Ballistic Pro Ladder
I've played BW for a while after a while (clearly EFR'd, muh skill). Seemed to me that machinegun spam (machineguns in general actually) is quite stronger than it used to be.
@Butcher: Data decay: sum kills, deaths, thaws, if it's greater than 3000 (atm) then divide it equally so the sum will be 3000. Inactivity decay: affects weighting ratio between your data and the average.
@Butcher: Data decay: sum kills, deaths, thaws, if it's greater than 3000 (atm) then divide it equally so the sum will be 3000. Inactivity decay: affects weighting ratio between your data and the average.
- iRobot
- Junk Administrator
- Posts: 3909
- Joined: Fri 06 Jan , 2012 10:37 am
- Contact:
Re: Ballistic Pro Ladder
Can you post the exact calculation?Izumo_CZ wrote:I've played BW for a while after a while (clearly EFR'd, muh skill). Seemed to me that machinegun spam (machineguns in general actually) is quite stronger than it used to be.
@Butcher: Data decay: sum kills, deaths, thaws, if it's greater than 3000 (atm) then divide it equally so the sum will be 3000. Inactivity decay: affects weighting ratio between your data and the average.
-
- Disappeared Administrator
- Posts: 4196
- Joined: Fri 19 Mar , 2010 1:21 am
- Location: Earth
- Contact:
Re: Ballistic Pro Ladder
The exact calculation:
When (kills + thaws + deaths) > 3000, then they are normalized by 3000 / (kills + thaws + deaths), so (kills + thaws + deaths) = 3000 (the data decay). Then your effi = (kills + thaws) / (kills + thaws + deaths) and weight = (kills + thaws + deaths) are calculated. After then, Bayesian estimate is made based on the amount of your data: ratio = activity * (1.0 * weight^2) / (weight^2 + avg_weight^2) + (0.0 (1 - weight)^2) / (weight^2 + avg_weight^2); this is now slightly scaled, so when weight = 3000 (with activity = 1.0) then ratio = 1.0. avg_weight is the weight of the average, which is now 500. Activity is between 1.0 and 0.0, and goes from 1.0 to 0.0 linearly starting from 14 days of inactivity and ending in 28 days of inactivity (on which the counter saturates). Then bayesian is calculated: bayes = ratio * effi + (1 - ratio) * avg_effi (avg_effi is the average efficiency of the server), this is in interval 1.0 - 0.0. And your skill is 10 * bayes, so we get the interval 10.0 - 0.0. It's actually quite simple math, only a bit more difficult to imagine on the first time. Thaws are normalized when converting to kills ofc (thawing ~ 2.5 players by proxy = 1 kill, or thawing 1.25 players by weapon (HMC etc.) = 1 kill).
When (kills + thaws + deaths) > 3000, then they are normalized by 3000 / (kills + thaws + deaths), so (kills + thaws + deaths) = 3000 (the data decay). Then your effi = (kills + thaws) / (kills + thaws + deaths) and weight = (kills + thaws + deaths) are calculated. After then, Bayesian estimate is made based on the amount of your data: ratio = activity * (1.0 * weight^2) / (weight^2 + avg_weight^2) + (0.0 (1 - weight)^2) / (weight^2 + avg_weight^2); this is now slightly scaled, so when weight = 3000 (with activity = 1.0) then ratio = 1.0. avg_weight is the weight of the average, which is now 500. Activity is between 1.0 and 0.0, and goes from 1.0 to 0.0 linearly starting from 14 days of inactivity and ending in 28 days of inactivity (on which the counter saturates). Then bayesian is calculated: bayes = ratio * effi + (1 - ratio) * avg_effi (avg_effi is the average efficiency of the server), this is in interval 1.0 - 0.0. And your skill is 10 * bayes, so we get the interval 10.0 - 0.0. It's actually quite simple math, only a bit more difficult to imagine on the first time. Thaws are normalized when converting to kills ofc (thawing ~ 2.5 players by proxy = 1 kill, or thawing 1.25 players by weapon (HMC etc.) = 1 kill).
- iRobot
- Junk Administrator
- Posts: 3909
- Joined: Fri 06 Jan , 2012 10:37 am
- Contact:
Re: Ballistic Pro Ladder
Makes complete sense, thanks
- Butcher
- V.I.P. Member
- Posts: 893
- Joined: Sat 15 Sep , 2012 1:31 pm
- Location: Germany
- Contact:
Re: Ballistic Pro Ladder
It does not sum up to me:
Lets see the first positions only:
Position Player Name Kills Deaths Thaw Points EFR Skill
1 newleaf*iRobot 21232 3977 1268.67 2 8.47
K+D+T=26477.67 then KDTnorm=3000/(K+D+T)=0.113303021
Normalized new values
KN=2405.649742
DN=450.6061145
TN=143.7441437
Then
Effi=(KN+TN)/DN=5.657699271-> (which is >1 so something fishy is here)
Actually the formula shall be:
effi = (kills + thaws) / weight and weight = (kills + thaws + deaths)
effi = 0.849797962
weight=3000
Activity =1 ->robot is everyday there
Anyway, with activity =1 then ratio=(weight^2) / (weight^2 + avg_weight^2)
if avg_weight=500 (which btw if you make the average with the 477 positions in the servers list gives: 1616.241332 ) then
ratio=0.972972973
Slightly scaled -> ratio=1
skill= 10*0.849797962= 8.49797962 -> weird cause I read in the page 8.47
So my questions are:
1) how much is avg_effi on the server?
2) how is this calculation change your skill respect to the old skill calculation? I don't see a difference besides it is only going to shown lower if your activity drops, but if you are active the skill is exactly the same value as before.
3) how it reflects if you have advanced?
When you are new noob then you have high count of deaths, low kills, low thaws.
When you are less noob then you have high count of deaths, some kills, some thaws.
When you are middle noob then you have medium of deaths, medium kills, medium thaws.
When you are experienced noob then you have medium to low deaths, high kills, high thaws.
When you are irobot noob then you have low deaths, high kills, and very low thaws.
For you to become experienced you must go though the process new->experienced which adds a lot of deaths to the count. But if you just use another cd key you may jump to experienced player and your skill is not weighted by all your history.
I think the skill system does NOT reflect your skill level, at least it does not for me. I believe the skill system should not be so weighted by the past experience.
Lets see the first positions only:
Position Player Name Kills Deaths Thaw Points EFR Skill
1 newleaf*iRobot 21232 3977 1268.67 2 8.47
Base 3000Izumo_CZ wrote:The exact calculation:
When (kills + thaws + deaths) > 3000, then they are normalized by 3000 / (kills + thaws + deaths), so (kills + thaws + deaths) = 3000 (the data decay). Then your effi = (kills + thaws) / deaths and weight = (kills + thaws + deaths) are calculated.
K+D+T=26477.67 then KDTnorm=3000/(K+D+T)=0.113303021
Normalized new values
KN=2405.649742
DN=450.6061145
TN=143.7441437
Then
Effi=(KN+TN)/DN=5.657699271-> (which is >1 so something fishy is here)
Actually the formula shall be:
effi = (kills + thaws) / weight and weight = (kills + thaws + deaths)
effi = 0.849797962
weight=3000
Activity =1 ->robot is everyday there
I think you mean->ratio = (activity * weight^2) / (weight^2 + avg_weight^2) + (1-activity)* (1 - weight)^2) / (weight^2 + avg_weight^2);Izumo_CZ wrote: After then, Bayesian estimate is made based on the amount of your data: ratio = activity * (1.0 * weight^2) / (weight^2 + avg_weight^2) + (0.0 (1 - weight)^2) / (weight^2 + avg_weight^2); this is now slightly scaled, so when weight = 3000 (with activity = 1.0) then ratio = 1.0. avg_weight is the weight of the average, which is now 500.
Anyway, with activity =1 then ratio=(weight^2) / (weight^2 + avg_weight^2)
if avg_weight=500 (which btw if you make the average with the 477 positions in the servers list gives: 1616.241332 ) then
ratio=0.972972973
Slightly scaled -> ratio=1
bayes=1*0.849797962+(1-1)*1= 0.849797962 (avg_effi not known assumed 1, anyway does not affect the calculation)Izumo_CZ wrote: Then bayesian is calculated: bayes = ratio * effi + (1 - ratio) * avg_effi (avg_effi is the average efficiency of the server), this is in interval 1.0 - 0.0. And your skill is 10 * bayes, so we get the interval 10.0 - 0.0. It's actually quite simple math, only a bit more difficult to imagine on the first time. Thaws are normalized when converting to kills ofc (thawing ~ 2.5 players by proxy = 1 kill, or thawing 1.25 players by weapon (HMC etc.) = 1 kill).
skill= 10*0.849797962= 8.49797962 -> weird cause I read in the page 8.47
So my questions are:
1) how much is avg_effi on the server?
2) how is this calculation change your skill respect to the old skill calculation? I don't see a difference besides it is only going to shown lower if your activity drops, but if you are active the skill is exactly the same value as before.
3) how it reflects if you have advanced?
When you are new noob then you have high count of deaths, low kills, low thaws.
When you are less noob then you have high count of deaths, some kills, some thaws.
When you are middle noob then you have medium of deaths, medium kills, medium thaws.
When you are experienced noob then you have medium to low deaths, high kills, high thaws.
When you are irobot noob then you have low deaths, high kills, and very low thaws.
For you to become experienced you must go though the process new->experienced which adds a lot of deaths to the count. But if you just use another cd key you may jump to experienced player and your skill is not weighted by all your history.
I think the skill system does NOT reflect your skill level, at least it does not for me. I believe the skill system should not be so weighted by the past experience.
"An BW match is a test of your skill against your opponents' luck."
-
- Disappeared Administrator
- Posts: 4196
- Joined: Fri 19 Mar , 2010 1:21 am
- Location: Earth
- Contact:
Re: Ballistic Pro Ladder
The webpage shows total kills / total deaths, not the current values (not updated to that yet ). Each time the current values sum exceeds 3000, it will reduce it to 3000, so historical data decay over newly collected data and will erase the initial deaths for unexperienced players. iRobot currently has: 2391.5 K, 458.8 D, 149,7 T. Indeed I had a mistake in the effi calc: (kills + thaws) / (kills + thaws + deaths), and it's just as you figured. Avg effi is 4.67 atm.
It's quite recent change, so it didn't kick very much yet.
It's quite recent change, so it didn't kick very much yet.
- iRobot
- Junk Administrator
- Posts: 3909
- Joined: Fri 06 Jan , 2012 10:37 am
- Contact:
Re: Ballistic Pro Ladder
Hey fuck you buddy, I have the 3rd highest thaws in the top 10When you are irobot noob then you have low deaths, high kills, and very low thaws.
Also I think thaw points are inflated by using defibs, I thaw everyone I see and I don't get anywhere near the 2k, 3k and 4k thaw point some people have, and it's not lack of games since I play everyday almost.
i have never once used the defibs, so I think they're messing up those figures
- Skaldy
- V.I.P. Member
- Posts: 264
- Joined: Tue 19 Mar , 2013 4:59 pm
- Location: UK
- Contact:
Re: Ballistic Pro Ladder
Hey - I'll have you know I've never used defib either
It's just that being top thawer means dying...frequently.
Perhaps I'll change my name to "The Sacrifice" or something.
It's just that being top thawer means dying...frequently.
Perhaps I'll change my name to "The Sacrifice" or something.
Last edited by Skaldy on Fri 26 Jul , 2013 4:40 pm, edited 1 time in total.
- Butcher
- V.I.P. Member
- Posts: 893
- Joined: Sat 15 Sep , 2012 1:31 pm
- Location: Germany
- Contact:
Re: Ballistic Pro Ladder
So lets assume I have :Izumo_CZ wrote:The webpage shows total kills / total deaths, not the current values (not updated to that yet ). Each time the current values sum exceeds 3000, it will reduce it to 3000, so historical data decay over newly collected data and will erase the initial deaths for unexperienced players. iRobot currently has: 2391.5 K, 458.8 D, 149,7 T. Indeed I had a mistake in the effi calc: (kills + thaws) / (kills + thaws + deaths), and it's just as you figured. Avg effi is 4.67 atm.
It's quite recent change, so it didn't kick very much yet.
K = 1500
D = 1000
T = 500
I calculate my Skill with this values and:
K D T effi weight ratio scaledratio bayes Skill
1500 1000 500 0.666666667 3000 0.972972973 1 0.666666667 6.666666667
And I play now a game and in average get in the game
K = 35
D = 20
T = 25
How do my skill values look like in the future?
Seems to me ok then if its what you say, only thing is that you will require to play 100 times to reach your average.. ie 3-4 months playing almost every day
"An BW match is a test of your skill against your opponents' luck."
Who is online
Users browsing this forum: No registered users and 11 guests