MVP in a timed race game

[lunch3]

With this new mvp system, it only makes sense to display the person who got mvp in the match info. Otherwise people by default will just think the most points spider got mvp. Heavensent catalogs mvps for our squad and this is only way he can show who got mvp factually.

[Force of Nature]

With this new mvp system, it only makes sense to display the person who got mvp in the match info. Otherwise people by default will just think the most points spider got mvp. Heavensent catalogs mvps for our squad and this is only way he can show who got mvp factually.

lnx is the guy who's incharge of TWD. he's the guy to talk too. However, formatting the MVP into the website will take some work. I would like to see the rest of the the stats go up as well since that would help me tweak the forumlas if I had a stored results of a lot of games somewhere.

[Jeansi]

What happens if shark average amount of reps left is 0???
cus then it would be 3/0 which doesn't quite work.

Not that it would ever happen but...

[Displaced]

What happens if shark average amount of reps left is 0???
cus then it would be 3/0 which doesn't quite work.

Not that it would ever happen but...

You don't have to worry about the divide by zero. I already thought about it and made it so (3/(repelleft)) evaluates to 4 if avgRepelleft is zero. So you get a bigger bonus.

yes im bored.

[Stabwound]

Using the !rating command, my avgRepelsLeft has always been 0, even though I do die on attach sometimes or get tked. I guess it rounds the number down?

[Kuukunen]

First of all, sorry for being a little late with this, I had lots of other stuff to do and I kind of forgot about it. Warning, the math terms might be a little off, as I didn't study maths in English.

Ok so if we want to do this the mathy way, let's do it properly.
I'd also like to add this suggestion is meant for the system that isn't gonna be implemented in a while. This is just a quick brainstorm of mine, it might not even work and might need some work, but at least it's something new...

We want to make best people MVPs right? And the best are the ones who help the team to win most. Well for the bot to know what helps the team most, it should first know which stats are important, and how important they are.

Let's say x is some number in player's statistics for that ship. Can be spid's jav kills, spid's deaths, etc.

Let's say w is the "win score" for that player that is currently being processed. It tells how well the game went for the whole team. This system is meant for time races, so it can be calculated in the following way: [(own team's time) - (enemy's time) + (time needed for win)]/[2 * (time needed for win)]
So w is a number from 0 to 1, which tells how well player's team did. If you don't believe me, examples:
win: 10-1 min: (10 - 1 + 10)/20 = 0.95
lose: 1-10 min: (1 - 10 + 10)/20 = 0.05
win: 10-9 min: (10 - 9 + 10)/20 = 0.55

Ok, next thing we do is check how well player's numbers correlate with w. This can be done by taking all x's from a set of games and calculating correlation coefficient for every pair of x and w.
I won't write how to calculate it here, but you can check http://mathworld.wolfram.com/Correla...efficient.html. (I'll use r instead of r^2) So basically r is number from -1 to 1 that tells how well x describes w. If r is 1 it means w is linearly proportional to x. (High x always means high w) Notice that it can be also negative. (For negative stats, like deaths.)

s is standard deviation for the set of x. (http://mathworld.wolfram.com/StandardDeviation.html)
m is mean for the set. (arithmetical mean that is)
so the score for each stat would be: (y is the stat from the game being processed) r * [(y - m) / s]
To get total score, sum scores from all stats.

So the learning part is that the bot stores in memory the x and w pair for all ships. So if there is 4 spids in each team, and every spid has 10 stats. 80 pairs will be stored, eight pairs for each stat.

You might notice the system will most likely be very biased for different ships. So I suggest all scores will also get an additional coefficient that takes amount of games in a ship, amount of mvps in a ship, total amount of games in every ship and total amount of mvps so that it would be evenly hard for every ship to get mvps. Too tired to think of a good formula right now, but that's not really a problem. I'd suggest this coefficent to other systems too. Some might claim that maybe only good players play terr so they really should get more mvps, but nah, I don't think that would be fair.

This system could be combined with the voting system (or some other system) and maybe the systems could fix problems in each others.

Problems:
*Biggest problem would be separating cause from effect. For example if one stat would be deaths in cram area: The winning team would defend a lot more, causing more deaths in cram area. Thus the system would interpret deaths in cram area is a good thing and would give lots score from it. Only good solution would be selecting stats that are not just a side product of winning. Or maybe separating stats with flag and without flag. Like DeatsWithFlagScore = deaths with flag / (flag time + 1 min). Where 1 min is just to prevent crazy numbers, if for example team holds flag for one second and someone dies in that second, the system would interpret that as equal to dying every second through the whole game.

*One problem might be the memory and CPU time needed. I mean correlation coefficients for 1000 games, with 80 stats, every single game, hmm... But with a little optimization this CAN be made so that every stat needs to be calculated only once. For example if the bot stores all the stats from one week before touching them. At the end of the week it calculates and stores the coefficients and standard deviations, and deletes all other information. Then uses the mean of the coefficients and such from a longer period of time.

*As noted before, revolutionary new playing styles might make these systems inaccurate, but if all coefficients more than X weeks old would be forgotten, the system would adjust to this too. (But I'd suggest stats that might take longer time to get accurate because there aren't much of them in one game, like teamkilling terrs, should be stored longer.)

*I don't know how much games this system would need to work properly. (If it even would work.) Maybe it's 100, maybe 1000 I don't know.


Something for all systems: I think terr kills in different areas should be counted separately, wb might easily get five terr kills outside base, but in base, where it really matters, it's a lot harder. (In base it SHOULD be rewarded with at least 10 times more points than from other ships.) Maybe outside base - tube, midbase and flag room.

Now after you've read this, (congrats, long post) you might think why see this much trouble just for getting MVPs a little more accurately? Well this is fun for me, but also I heard the scoring system could have some more important use than just the MVPs in base. And actually for me, the good old score based MVP was fun too. Yeah, terrs might not get as many MVPs as spids, but when they do get MVP it's something special.

[lunch3]

For shark mvp, it might be very helpful to add following variable.

The average length of time the shark stays alive.

Say that for an experienced shark a life of 6 seconds is expected, if they use 3 reps. If the shark dies sooner than 6 seconds that hurts team and means shark repeled too fast not conserving them, or shark wasn't able to use all reps. But if the shark is alive much longer, then that usually means the shark is fucking around, not at heart of battle, and probably waiting to accumulate energy to lob bomb or mine somewhere. So denote it as (SLT) Shark Life Time.

Then add it into equation as some term that gives the highest ratings when SLT/6=1. When above or below 1 this term should hurt rating proportionality. Dunno what Math term to use after thinking about it for awhile though.

Also think the jav and wb formula should be increased a little to make it more likely that those ships get mvp.

[Sarien]

Suggestion for Sharks:
Since you talked about the radius around the Terrier:
What about a Radius around a shark the diameter of the area of
effect of a repel, and since you can mark entrances and exits,
you could determine how many repels have been used to shove
enemies back out of the entrance. If you could track bombs,
You could tell how many bombs a repel has tossed away from the FR,
thereby saving people. IE Doing well as a shark.

[Force of Nature]

Suggestion for Sharks:
Since you talked about the radius around the Terrier:
What about a Radius around a shark the diameter of the area of
effect of a repel, and since you can mark entrances and exits,
you could determine how many repels have been used to shove
enemies back out of the entrance. If you could track bombs,
You could tell how many bombs a repel has tossed away from the FR,
thereby saving people. IE Doing well as a shark.

This is quite hard to implement but I think it can be done. However it would also be resource intensive compared to other stats

[Stabwound]

There is really no way a bot can watch and judge a shark or terrier's style, which is the only way to determine if they should be MVP. I know that the bot can really only do things with countable stats, but the terrier and shark really should not get an MVP from points or kills at all. For a shark to get MVP, you really have to mine a lot more than you really should. Even if you get a kill and a teamkill at the same time, the teamkill hardly offsets the amount of rating points you get for the kill.

Stuff like Sarian suggested would only promote a certain type of play (IE cramming) even more, just like how the current system rewards wreckless mining. Recording the time a shark stays alive just forces them to play quicker or slower or whatever. Depending on the situation, you could be dead in 3 seconds or alive for 2 minutes. A shark shouldn't get their MVP taken away just because they stay alive for 120 seconds because the enemy terr dies 5 times in a row.

Yesterday I got shark MVP by mining EVERYWHERE. I had over 25 teamkills and my team lost, yet I still got MVP.

[Force of Nature]

Changes made to fix of rating system:

warbird: .6Points * (.07wb + .07jav + .05spid + 0.12terr + .05x + .06lanc + .08shark - .04deaths)

jav: .5Points * (.05wb + .05jav + .06spid + 0.12terr + .07x + .05lanc + .09shark - .05deaths - (.07wbTK + .07javTK + .06spiderTK + .15terrTK + .06WeaselTK + .07LancTK + .09SharkTK))

spiders: .4points * (.06wb + .06jav + .04spid + .08terr + .05x + .05lanc + .09shark - .05deaths)

terr: .8points * (.06wb + .06jav + .08spid + .12terr + .1x + .06lanc + .09shark - .12deaths)

weasel: .8points * (sum(.09allships) - 0.05deaths)

lanc: .6Points * (.07wb + .07jav + .05spid + 0.12terr + .05x + .06lanc + .08shark - .04deaths)

shark: 1.2points * (.12terr + sum(.09allotherships) - 0.005deaths - (.1(allothershipstks) + .15terrtk + .11sharkTK))

[(.4 * (3 / avgRepelLeft)) doesnt work]

avgRepelLeft isn't working. Atm its only giving everyone a value of 0 no matter how many repels they use. It never worked. I'll look into the implementation of this and either fix it or give a differnt bonus to the shark.
(I have shark coverage time in mind - where the shark gets bonus rating proportianal to how much time there is a shark alive to cover the team as long as its got one repel and not dead together - the feasablity of this still needs to be thought out)

!rating <player name> no longer displays the current MVP. It still displays the rating of the player requested and also the formula used to calculate the rating (if the player is in the game !rating is enough and will ensure that his stats are shown)

!mvp displays the current mvp in the game.
Later this is going to list the top three players ratings which isn't implemented yet.

ship changes no longer get MVP. Infact the way its implemented they get a 100000 rating deduction. This is to ensure that if everyone changed ship then mvp would be alteast given to someone resembling sane stats.

[Kuukunen]

ship changes no longer get MVP. Infact the way its implemented they get a 100000 rating deduction. This is to ensure that if everyone changed ship then mvp would be alteast given to someone resembling sane stats.

I think the simplest and fairest way to determine mvp when ship changes are involved would be to judge stats by the factors of the ship they're done in and then if ships are changed, they're added together. From the view point of the programmer would be something like:

Every time a ship is changed, the stats are added to temporary score variable using the factors of the ship player is changed from, (0 in the beginning) and then all stats are set to 0.

[Noah]

I like base cuz it's not all tied up in stats.. It's freedom in a sense, prevents choking.

[za gophar]

THIS IS BULL SHIT. quit giving terr and shark mvp. anyone who takes MVP seriously when a SHARK gets it is fucking out of their mind. this is encouraging sharks not to do their main job, which is repel for the team. ill be sure to remember this next time i get shark so i can mine like fucking crazy and get mvp.

SECOND, this is a highly subjective system which is therefore highly subject to error. why is more value placed on a spider killing a shark than a spider killing a terr? bull shit. some ships worth more than others? bull shit. if some all knowing being made that system, i wouldnt mind so much. but a person made it, and his opinions about how to base could be and probably are different from mine. so are we all supposed to play like he has indirectly told us how to?

THIRD, the terr ranking is bull shit because how well a terr does is at least 50-60 percent how much his team helps him. it is bull shit that a terr could just sit in the cram the whole game while his team takes care of business in midbase, crushing the other team. thats a free ride to the mvp, while things that a terr does that are truely outstanding are not taken into account because they have no statistical measure.

Ok so if you are going to fix something that is not broken, at least keep it objective and unflawed. the system has many large flaws, you can't deny it. please tell me that no one takes the current MVP system seriously.

[Stabwound]

Yes the current MVP system is pretty damn bad. No offense, but obviously whoever made these new formulas isn't a hardcore baser, because it is very stupid that killing a shark gets you way more points than killing a spider.

Also agreed with the shark thing. I usually play shark and if I know I have a chance at getting the lovely MVP, I put as many mines as possible. I think if you get around 20 real kills in shark with the flag, you will likely get MVP.

The only way a terrier can get MVP is if their team is good enough to allow them. I think a terr should have to have under 5 deaths to even be considered for MVP.