PDA

View Full Version : small suggestion for s3


zz-
08-03-2011, 10:26 PM
For season 3 could you make it so a W-streak or L-streak doesn't magnify both possible outcomes of every game? It doesn't make mathematical sense to go 12-4, then lose 4 games and be back at +0 for the session despite going 12-8 because the losses were amplified due to your upward trend.

Conversely, a player that starts 3-13 shouldn't shoot higher upwards after a win simply because he is trending downward so hard. In fact, it should be the opposite.

The point of the amplifier is to help trending players reach their destination sooner. Amplifying each type of outcome after each type of streak prevents the very thing it's put in for from happening. At the moment, a win streak only gives you a (nearly) coinflip's chance of the amplifier helping you, and a coinflip's chance of it hurting you -- rather than simply helping you trend harder in the direction you're currently trending. What I'm saying is, if you're trending up, wins should be amplified and losses should be held equal. And likewise for losses/trending downward.

This will help new players reach their appropriate ratings much faster since it will prevent so much of the skew that occurs from winning while you are losing overall, and vice versa. Also will make for more possible lead changes at the top, etc etc. Good all around

Thoughts? Perhaps the amplifier could apply to each outcome, but more so for the direction you're moving?

Ribilla
08-03-2011, 11:09 PM
My understanding is, it isn't done on streak, it's done on net points gained in the last x amount of games.

Say you are at 1000 rating, you have a 10W streak and shoot up to 1500, you are very volatile at this point. You then lose 5 games in a row, taking you down to 1200. You have only varied by 200 points in the last 15 games, so you are now non-volatile, i.e. not on a streak, this means that even if you continue to lose, you will lose very few points, as you are tending back to your original rank. The next 5L would probably only cost you the 200 points to take you back to where you were.

nobodyhome
08-04-2011, 12:36 AM
this doesn't quite work. the multiplier is less of a "streak multiplier" and more of an "uncertainty multiplier" that increases the point value of each of your games the more it thinks it's uncertain that your rank is correct. thus when you are "streaking" then your rank is obviously not correct so it increases the value of each game.

mathematically, you'd want the multiplier to pull your gains as well as your losses because you want to preserve the property that each game, if you are indeed ranked correctly, has an expected value of 0 points of change. if you multiply one way but not the other then you'd mess this up.

more generally, what would happen if the multiplier only worked one way is that once you get on a streak, you'd tend to overshoot your correct ranking, and then take longer to go back downwards. with the two-way multiplier you do tend to lose a lot per game once your streak ends, but that's because the streak multiplier overshot your ranking anyways.

TRUEPAiN
08-04-2011, 02:06 AM
Alsssooo.. custom_random should just play through a rotation.. either that or just make a map poll vote for next map?

cause random, not so random.

Much like shuffle on the ipod when it was first released.

Dark_Sage
08-04-2011, 02:24 AM
with the two-way multiplier you do tend to lose a lot per game once your streak ends, but that's because the streak multiplier overshot your ranking anyways.

The problem with this is when you lose a game that you had a 40% chance of winning, and lose 40 points whereas your teammates can lose just 15. You're rating is obviously still correct if you lose a game with a crap team, so why lose way more points? Likewise, if on a winning streak why gain twice as many points when winning a 60% chance game? I don't like the modifier myself, as having both that and not having all games be 50-50 points wise seems counter-intuitive.

And I 100% agree with tpain. No-one would want to play woods or hardcourt if given the choice.

elxir
08-04-2011, 02:48 AM
i want to play hardcourt and woods...:confused:

Carlos98
08-04-2011, 04:26 AM
i want to play hardcourt and woods...:confused:
10charssssss

naethy
08-04-2011, 06:46 AM
Another couple (very small) suggestions:

1) Remove "avg leaves" and "avg ping kicks" from the map page, I don't see them having any practical use. Replacement idea: statistical deviation of the team's players' ranking. This would allow one to see if the team with wide spread in rankings fare better, worse, or the same as teams that are more "even". Other ideas welcome.

2) Draw out goals/assists per match to 2 decimal places. After 100 or 200 games these get pretty cemented, and it's hard to see any change without doing the stats on your own.

3) See win/loss for each map. Would LOVE to know which maps I suck at, though my guess is ice, hardcourt, and core.

4) It would be REALLY cool if you could tell more easily how good you are with each plane: for instance, I have a hunch that I'm a much better loopy than whale, but do to a dearth of good whales whenever I'm on, I often play whale instead. I wonder what my "whale rating" and "loopy rating" are individually. I realize that this idea is probably BAT**** INSANE COMPLICATED to impliment, but would be awesome if it worked.

5) Add more maps to rotation this season please :-) I personally would love to see ball_decimal played more, I think it has a lot of potential.

And remove ball_ice. EVERYTHING IS A PASSAGEWAY AND I NEVER GET WALLS.

--slight

Tekn0
08-04-2011, 01:59 PM
Alsssooo.. custom_random should just play through a rotation.. either that or just make a map poll vote for next map?

cause random, not so random.

Much like shuffle on the ipod when it was first released.

This. chars!

We are palying alternate games spectating one in between. At least make the random skip last 2 maps instead of just one.

nobodyhome
08-04-2011, 08:42 PM
The problem with this is when you lose a game that you had a 40% chance of winning, and lose 40 points whereas your teammates can lose just 15. You're rating is obviously still correct if you lose a game with a crap team, so why lose way more points? Likewise, if on a winning streak why gain twice as many points when winning a 60% chance game? I don't like the modifier myself, as having both that and not having all games be 50-50 points wise seems counter-intuitive.

And I 100% agree with tpain. No-one would want to play woods or hardcourt if given the choice.

If you had won that game you would've won 60 points, thus putting your expected value of the game to still be 0 points total (60 * 40% + (-40) * 60% = 0). The ladder system is not able singular results but trends over time.

If we remove the multiplier system we get what we had last season, which nearly everyone agrees was a crap system. Just think about how many times having an overrated player on your team meant nearly an auto-loss, and how often it occurred. Now think about how often you had an overrated player on your team this season.

nobodyhome
08-04-2011, 08:44 PM
Alsssooo.. custom_random should just play through a rotation.. either that or just make a map poll vote for next map?

cause random, not so random.

Much like shuffle on the ipod when it was first released.

Random IS random (with the exception of the cannot the same map twice in the row rule). What you are asking for me is to implement not a random rotation, but an even rotation (which can be done, seeing as that's what most people seem to want when they say they want "random").

That's the same deal with the ipod. Shuffling originally used to be completely random, which everyone hated. They had to tweak it so it skewed the probabilities toward songs that weren't played as often, which by definition is not random.

nobodyhome
08-04-2011, 08:47 PM
Another couple (very small) suggestions:

1) Remove "avg leaves" and "avg ping kicks" from the map page, I don't see them having any practical use. Replacement idea: statistical deviation of the team's players' ranking. This would allow one to see if the team with wide spread in rankings fare better, worse, or the same as teams that are more "even". Other ideas welcome.

2) Draw out goals/assists per match to 2 decimal places. After 100 or 200 games these get pretty cemented, and it's hard to see any change without doing the stats on your own.

3) See win/loss for each map. Would LOVE to know which maps I suck at, though my guess is ice, hardcourt, and core.

4) It would be REALLY cool if you could tell more easily how good you are with each plane: for instance, I have a hunch that I'm a much better loopy than whale, but do to a dearth of good whales whenever I'm on, I often play whale instead. I wonder what my "whale rating" and "loopy rating" are individually. I realize that this idea is probably BAT**** INSANE COMPLICATED to impliment, but would be awesome if it worked.

5) Add more maps to rotation this season please :-) I personally would love to see ball_decimal played more, I think it has a lot of potential.

And remove ball_ice. EVERYTHING IS A PASSAGEWAY AND I NEVER GET WALLS.

--slight


All good ideas (no comment on the maps suggestions though--I don't handle maps, ask lix). Adding standard deviation of teams to the website could be done easily (although I wouldn't want to remove ping kicks and leaves--although fairly useless for most people there's no point in removing information, and I know some admins use it). The "figure-out-average-point-gain/loss-for-each-of-your-planes" idea gets complicated when you take into account that you can switch planes mid-game, but it can be done with weighing the point values by how much percent of the game you played it with.

elxir
08-04-2011, 09:57 PM
I've not heard anyone voice negative opinions of ice since like, it was first introduced...

Decimal could definitely be added to the rotation, ball_pulse and ball_hydro are close...tbd_leaf and tbd_metropolis are potentials as well

they all need tweaking and testing, though...which none of you are willing to do in custom_tests.

[hlx] FatTristan
08-04-2011, 11:48 PM
I feel like the rating system needs a tune-up as well. I think it should ignore how much of a streak you are on altogether. How many points gained or lost in a certain game should be based only on team win percentage. It doesn't make sense to want to revert a losing or winning player. Sometimes, weird stuff happens.

nobodyhome
08-04-2011, 11:52 PM
There's nothing wrong with synnik gaining 350+ points over three games. See my above explanation for how ladder rating system works based on expected value over time rather than individual wins and losses. The fact that synnik was able to continue to win a third game after shooting up 200 points just attested to how underrated he was.

If this had occurred in season 1 (where the multipliers did not exist) then what would've instead happened is that having synnik on your team would've been a near auto-win for at least 8 games, incorrectly skewing the point gains and losses of everybody else in those 8 games.

[hlx] FatTristan
08-04-2011, 11:53 PM
(This one's bigger)
Not to downplay his ability to win these tough games, this is just strange.

sunshineduck
08-04-2011, 11:54 PM
inactivity bonus to the uncertainty multiplier

[hlx] FatTristan
08-05-2011, 12:00 AM
If this had occurred in season 1 (where the multipliers did not exist) then what would've instead happened is that having synnik on your team would've been a near auto-win for at least 8 games, incorrectly skewing the point gains and losses of everybody else in those 8 games.

Aw damn, that's true.
I guess your solution is the best way to do it. It balances number of games with win percentage.

I was about to further suggest that people with higher ratings gain less and people with lower ratings gain more, but that doesn't benefit players who play a lot of games and only favors win percentage.

naethy
08-05-2011, 03:16 PM
Adding standard deviation of teams to the website could be done easily (although I wouldn't want to remove ping kicks and leaves--although fairly useless for most people there's no point in removing information, and I know some admins use it).

I should clarify: I don't suggest removing pings kicks and leaves from PLAYER pages, just from MATCH pages, since for every game I've ever seen it's always 0% or 1% and does not predict the outcome of the match at all. That free space could be used for the statistical deviation thingy.