MVP in a timed race game

[Randedl]

in hz they do it by rating, with every possible stat being worth a rating point. might want to make it ship specific here though. . . maybe:

spiders: points * .8kills/deaths * (1 + .125x + .01y)
x = terr kills
y = shark kills

warbird: points * (1 + x/y + .5z)
x = terr kills + shark kills
y = kills
z = terr kills without flag

terr: points * .5(x/y)
x = total cumulative bounty - 3y
y = lives

jav: points * (1 + .05x - .1y) * (1 + .z)
x = kills
y = teamkills
z = terr kills

shark: team points/8 * (1 + .02x - .05y + .1z) * (1 + x/a)
x = kills
y = teamkills
z = terr kills
a = deaths

i just made up random numbers, but most of that takes into account the various functions of ships maybe? just do trial and error with real stats to see how it would work.

[Desperate]

BUT BUT BUT BUT I want a good stat that would be easy for sharks to get.

What about most teamkills?

If Jeromes idea would be used, I don't like the pming idea. Just message it like earlier.

[Stabwound]

That's not what I would call a "good" stat.

[Jerome Scuggs]

Well, i guess it could spam arena, but then you have the stream of "gg"s which then push the arena stats all the way to the top. So iguess it could spam it, then do the !stats on request.

OROROROROROROROROOROROROROROROROROROROROROROR.

make a website like elim results, and have them post it there

[Force of Nature]

in hz they do it by rating, with every possible stat being worth a rating point. might want to make it ship specific here though. . . maybe:

spiders: points * .8kills/deaths * (1 + .125x + .01y)
x = terr kills
y = shark kills

warbird: points * (1 + x/y + .5z)
x = terr kills + shark kills
y = kills
z = terr kills without flag

terr: points * .5(x/y)
x = total cumulative bounty - 3y
y = lives

jav: points * (1 + .05x - .1y) * (1 + .z)
x = kills
y = teamkills
z = terr kills

shark: team points/8 * (1 + .02x - .05y + .1z) * (1 + x/a)
x = kills
y = teamkills
z = terr kills
a = deaths

i just made up random numbers, but most of that takes into account the various functions of ships maybe? just do trial and error with real stats to see how it would work.

This will be the temporary fix to the timed race mvp. Since it seems relatively easy to do. I"m not going to bother changing the numbers. But for the permanent project here is what I have in store. I expect it to take about three months so its not meant for TWL but for the betterment of the MVP system itself if more people use it.

I think I'll make the MVP system autonomous. That means its going to fix the weights of each state itself, sorta like an AI learning after each base game about what the new trend of MVPs is. How does it decide?

The concept is as such. It will have a modified voting system. It will rate players according to its rating system given weights it has predefined. At the end of the game. Its going to ask for a 1 min vote. The vote however doesn't define the MVP. It just helps the bot learn what people think are important. And the bot uses this information to refine its weights to better estimate who is MVP next games.

For example: each ship is going to be rated on various stats. Each of these stats are going to have a weight ranging from 1 to 0 or 0 to -1 depending whether it is a positive or a negative attribute in the increments of 0.001.

Ok so lets say the stats that are recorded for the Jav is such
Jav
#kills with flag = x1
#kills without flag = x2
#total kills = (x1 + x2) = x3
#deaths = x4
ratio = x3/x4 = x5
#teamkills = x6
#points = x7
#type of ship killed - wb = x7
#type of ship killed - jav = x8
#type of ship killed - spider = x9
#type of ship killed - terr = x10
#type of ship killed - weasel = x11
#type of ship killed - lanc = x12
#type of ship killed - shark = x13
#multikills without tks = x14

So, the jav has 14 weighted stats all which cannot exceed 7 (half the number of total stats - this is because if all the set to 0.5 they add upto 7). The negative stats change it so for the jav it would be 6 because teamkills is a negative stat and would subtract 0.5 weight. If one of the stat increase the other will need to be adjusted so taht they don't exceed 6. So, if one increases the other decreases. Over time this will show which stats are truely important in the players minds.

Each ship will have a set of stats as such and will have their own wieghts.

There is also going to be some reference stats which are obtained from good players at the start. For the purposes of the equations these are labelled z1..z14.

If a1 --- a14 are the weights of all the stats. Then the rating will be
1000 * [(a1*x1/z1 + a2*x2/z2 + a3*x3/z3 .... a14*x14/z14) / total number of stats (14)].

This will be done for every ship and the best rating will be MVP.

Now how does the bot learn:

Well after the game. The stats will be displayed and the players will be asked to vote for one min. If no one votes then the top rating person is MVP. Only the first or the second person can be MVP depending on how many votes they make up for the difference.

Let me illustrate.

Say 1st player has a rating of 700 and second player has a rating of 750. (these numbers aren't set in stone and will be revised while trying out) Then every 10 rating points will be a vote. The second player can only get MVP if he has 5 more votes than the first player.

However, you can vote for anyone you like. If there are more votes to another player than the first or second rated players. Then given there are a minimum of atleast 3 votes. The bot will adjust the weights for the ship type of the voted player so next game it will be easier for that ship type to get a higher rating. This way I can kinda simulate the changes of ship uses in basing matches.

The con of this system. It will take a while to get the bot to learn the norm and have proper weights but once it does it should be able to predict MVP quite accurately.

[Force of Nature]

I'm just writing this so i can access this from anywhere. Those interested read.

Ok how are the weights changed by voting.

When voting you have the option of selecting what stats you think are important if you want. If you don't, then the current stats of that player are cross-referrenced to a set of stats which I will gather through good players playing base matches that are mvps. If some areas of the current stats exceed the predefined stats the predefined stats are adjusted. Also, the weight is incremented.

weight which is a1 is incremented by a factor of 0.001 * number of people who voted more than the top rating player * exponential function.

a1 = a1 + 0.001 * expFunc(#peoplevotedforship-#peoplefor1stperson).

The reason behind the exponential function is that the increments are traveresed faster. If there is a landslide on voting for a single person the increments will be incremented faster via the exponential function. This way it takes less games for paradigm shifts in base games for the bot to recognize the new weights.

Now the rest of the weights have to fixed so they dont exceed the weight limit for that ship. As shown above, the jav is fixed at 6. so an increase in one stat will decrease other stats. All the stats not mentioned by players and the stats that are lower than the predefined stats will decrease by a propotional amount which they are normally. Therefore a low stat will go down faster than a highly weighted stat. The total of the new weights will again add up to 6. There will also be a max of 1 so that a stat cannont exceed one no matter how many times its voted. this is to ensure that a particular stat isn't blown outof proportion.

[Jeansi]

I have no problem with the current most points system for mvp but voting sounded like a good idea for twl-b games and so. For ?go base it would be a waste of time I think.

Didn't have the time to read FoNs last posts though but a system where all ships have an equal chance to get the mvp sounds like fun.

[Randedl]

fon, i'd be happy to help out with some of the formulas if you want. i'm interested.

[za gophar]

I still say the spiders deserve it. they are the most under-recognized ship, and spider MVP is a good way for people new to the arena to gain recognition.

[RabbitRapist]

the most practical way to do it is to vote

another option would be to compile stats that are typical of a very good game in each ship. maybe 200-0 in spid. 100-0 in terr, 20-150 in shark. and whatever else you can think of. then get the bot to calculate the difference between the perfect game and the game the player had. so you'd measure it in %-difference-from-perfection

just a thought

[lunch3]

I like the idea of using equations and weighted numbers to determine mvp. That is the most fair way, but certain things are hard to measure, like maybe the more reps a shark can do in one game gives them more weight for mvp then sharks that die before using all their reps, is there any way a bot can count or measure this stuff? Can the bot also recognize tds opposed from regular kills? If it could count such stuff even better....otherwise some of them can be approximated somehow.

[Adman]

the last TWL bot knew terrkills and teamkills.

[Due]

the last TWL bot knew terrkills and teamkills.

not terrkills (i think)

[Stabwound]

Yes, IT DID INFACT record terrkills!

[Jerome Scuggs]

no your recorded them, its hardcoded into your system

the matchbot then tapped your database and retrived the info