Jump to content

Aperson2

Mathematics Contributor
  • Content Count

    21
  • Joined

  • Last visited

About Aperson2

  • Rank
    Post Derp Rage Poster

Profile Information

  • Gender
    Not Telling
  • Server
    NA
  1. Updated Main Post This system works for all tiers but in the end it is comparing all the tanks to tier 10 tank (used as a baseline for measuring "skill") so basically as a result the system defines value in this ratting system based on how well you would do in tier 10 tanks. it should work well even at lower tiers beause you are looking only at recent games played in the low tier tank and comparing it to that same players recent games in tier 10 tanks so automatically players who don't have tier 10s are not used in the system resulting in the low tier values being largely based in "sealc
  2. Thanks, really usefull, btw i am still curious on what you mean by your last comment, just didn't get what you were saying
  3. yep. btw how was this data collected, aproximate number of player and range of time. Don't get exactly what you mean by that. Mostly i don't get exactly what you mean by account values.
  4. The Basic Idea: Use relationships between different tanks to covert stats such as dmg or winrate to what people who get those stats typically get in tier 10s(averaged across all tier 10s that aren't premiums). Interval data would be used to construct the relationships so people's skill level changing over time shouldn't be an issue. Details: A linear fit using interval data is used forcing it to go through (0,0) also note the orthogonal distance from line to points should be minimized not the vertical distance so the relationship between tank1 and tank 2 should be the same as the inverse
  5. I tried averaging the defense both sides using the same weights for each player and and the results aren't too good from just cross checking stuff so definitely need a larger sample size to may by get anywhere. by the way can you by any chance give me your interval data, would save me quite a bit of waiting. As for spotting and other stuff i am getting reasonably good numbers for the sample size as a check i went from tank -> tank 2 -> tank 3 and compared that to going from tank 1 -> tank 3 and if the data is good the two numbers should be close so i got 1.800127886446354
  6. Ok i got the interval data from top 300K player on EU server, and here are some of the graphs that resulted. The time i waited is less than a week(collection took some time making the effective time difference between the two times i collected the data lower) so on most of the graphs the average number of battles per player is very low(# of battles = Min(battles in first tank, Battles in second tank)) though given the number of players and 34 comparisons being made(one per T10) the variance should work it self out for some of the categories at least on the top of the graphs it says somethi
  7. UPDATE: I have been busy for the last week or so but i have had time to set up data collection of the top 300K players on the EU server to get data so i can measure change over time. Unfortunately i ran into some ram problems and a bug in my code and ended up wasting quite a bit of time. So i should get a good version in a week or two that uses data based on recent data if the data is good enough after that amount of time. Can't reply to comments, computer working on collecting data leaving little room to run anything else since because of the RAM problems
  8. Ty for the info guys. I am not using the same formulas as before but the example you mentioned the problem stems from 2 things first and harder to deal with, the correlation between the how well you do in tier 4 and how well you do at tier 10 isn't very strong though it gets stronger as you go up the tiers(hence the lower weight given to lower tiers) and this is resulting in flatter best fit lines that aren't really that accurate at the lower tiers. The second reason for the problem is of course that the data is from the top players by battles played and not exactly the best way to repre
  9. If you know anyone(or are someone) who has low tier tanks that they seal club in please tell me, I would like to see if that results in inflated stats (seal clubbing tanks that aren't normally seal clubbed in are especially good). Feel free to PM me if you don't want to post publicly.
  10. Please tell me if you see something that seems off in the ratings per vehicle The overall is weighed based on the tier and number of games played in each tank so your low tier tank's ratings should have little affect on your overall (100 tier 1 games have same value as 1 tier 10 game) Had to split the replies up because it didn't work as a single response 1706.79 2478.27 1721.91 1251.51 Can't changing it too much on a regular basis and this is just for testing, will try to get it integrated into a website when i am done so it can be seen
  11. Version 0.3 (stating to number them now to avoid confusion): I am back and have a basically finished version i think... It relies on a WN8ish system for the final conversion from tier 10 values into a single number so the final numbers should correspond reasonably well to your WN8 (but be bit lower since its based in tier 10s) Upsides when compared to WN8: LTs and stuff don't get extreme values (padding at least like this shouldn't be a problem) ---- My 4th best tank by WN8 is a LT but in this the LTs are not near the top Tanks that have a smaller distribution of performance shouldn'
  12. Out of town for a while, will probably be home tomorrow, Can't do any thing without my desktop
  13. For those of you who want the Functions used to find the different values https://drive.google.com/file/d/0B4Kgy_Y2LJJ2UEJJUHJ4dk5SWGc/view?usp=sharing Its a text file containing all the functions for the different tanks that is organized like this (the functions are just the line of best fit between that tank and all the tier 10s added together) {Function for Spotted Value(j = Spotted value for this tank), for wins, dmg, frags, defence} the code takes the avrage dmg value for example on a T-34 puts it into the dmg function for the T-34 gets a huge number(about 34X what you
  14. So the number of battles is taken into account: if you have 2X the battles in a tank it will have 2X the affect in the overall rating( the tank based rating is just based on your average over those battles) Also the tier is taken into account by tier^2 so a tier 1 has 1/100 the effect as a tier 10(might change the exact number on this) Forgot to include that in the post sorry as for the not disclosing that much stuff, its basically because i am changing it too much mostly trying to figure out a good models i can use and where stuff like correlation coefficient of the data should b
  15. Wow i did such a terrible job of explaining my self lol. So It is a rating system based on the prediction system so it isn't saying you can get 3K DPG in a T-70, that is just a rating for your T-70kind of like your WN8 for your T-70(tier is taken into account when calculating the overall rating so your T-70 has almost no effect on your overall) So how its calculated, I didn't exactly say because i was changing it a lot as i went along but for now i am using this: First just find a line of best fit between a tier 10 and the tank who's rating you are tiring to determine so you get this
×
×
  • Create New...