|
Just FYI re-reading one of your previous responses to me, if the constant were appropriately adjusted downward from previous rounds it would be about 3.9 (a 14.6% decrease in the constant to reflect the 14.6% decrease in total points, assuming the same number of streak points and ace matches.) Assuming the constant was accurate in the first place (which it wasn't) and rounding up to 4 -- both jaedong and bisu would be worth 8 points under the new scoring system with the adjusted constant. Flash would be worth 14 and a half.
But since it was inaccurate, let's examine the previous constant in light of the previous round's scoring in order to establish a new constant by subtracting 14.6% from an accurate old constant. Under the old scoring, if we started with a definition of 1-pointers as players that will lose all games they appear in, the score of a 1-pointer would be equivalent to the number of team wins per match. Since each match is zero-sum the average team wins for all teams in 11 matches is 5.5. which would've been an accurate minimum starting point since under the old scoring there is no way to drop below your team wins in score. (The average would be very close to, but not exactly 5.5 -- that not all teams have the same number of players skews the average very slightly in favor of those teams who have more players. Last round this skew was only +8 points from the ideal average of 814 or 1%, which would add + .05 to the constant.)
We have one more thing to address in the old scoring, though, and that's the lineup appearance points. 1-pointers who did not play but appeared in the lineup would earn one more point than a 1-pointer who appeared in the lineup and played and (by previous definition) lost. Now, assuming that there weren't a ridiculously high number of 3-0 games last season, which I think is a pretty safe assumption, we can use the results from last season as a baseline for an appearance point model. Last season 528 points for lineup appearances were doled out, though only 490 lineup appearances actually happened. Considering the distribution of lineup appearances for 1-pointers, 103 of 528 lineup appearances were for 1-pointers. Thus (38 * 103/528)/490 will get our % increase for our expected point gain for 1-pointers under the old system relative to our current 5.5 +- .05 constant. That comes out to .015 or 1.51% or ~.08, which, rounded up, brings the constant to a significantly more accurate 5.6.
Adjust 5.6 downward by 14.6% (recall, this is to reflect the overall % point decrease expected with the scoring changes) and we get ~4.8 for our new constant.
Let's see how much various players would cost with the new scoring and the new constant, rounded to the nearest integer:
+ Show Spoiler [list] +
Flash 12
Zero 11
Sea 10
Effort 10
Violet 9
Really 8
Luxury 7
Jaedong 7
Bisu 7
Free 7
Calm 6
Fantasy 6
Roro 5
Ruby 4
Kal 4
Leta 4
etc.
As you can see, even with an accurate constant, the point distribution is heavily skewed toward players with higher win percentages as opposed to 50/50. Under old scoring, 11 win flash scores and costs double to a 6 win ruby. Under new scoring, 11-1 flash scores and costs 3.41 times (12.08 to 3.54) as much as 6-7 ruby, who in turn scores 8 points less than a 4-0 action.
I don't think that's a change for the better. At all. You might disagree. I think it's fine to play it out for a season, I've already made my best guess as to who will have the best win% this round, plus it's kind of late to change the scoring again.
The constant does need to be changed though. I haven't even gone into the effect of an inaccurate constant in this post, hopefully it's just already obvious why having a constant that is too high will significantly deflate player costs and trade values midseason.
I would encourage you to check my math in case I made a mistake or overlooked something, but I'm pretty sure I didn't.
|
Hi. Sorry if this question has been answered before.
How do the player prices affect the player trade values after the first week? What I mean is: are the trade values determined solely based on the player price + first week performance?
Or are the trade values based on the player's trade values at the end of previous seasons? In light of the posts directly above, it seems it might take into account two previous seasons (or is this just the player price?), in which case how are we to determine what a player's actual trade value is at the start?
I'm asking this question because of this: for example, Sea and Zero are both priced at 9. However at the end of season 5 their trade values are 9.87 and 10.53.
If I had Zero and he performed slightly worse than Sea in week 1:
- if trade values after first week were based solely on price - then I can't trade out Zero to get Sea; or
- if trade values were based on something else, say FPL round 5 trading values - then I might be able to trade depending on the numbers.
Hope it wasn't a silly question with an obvious answer. Thanks!
|
Just because we're not agreeing with your opinion does not mean we're not "addressing fantasy with the same level of thought" as you.
C'mon man, obviously you're not, you didn't even notice that overall scoring was going to be decreased this round when you adjusted the scoring rules. As a result, you guys got the constant wrong too, and now you're adjusting costs upward to deal with the deflation caused by your inaccurate constant. And now you're arguing with me with irrelevant conclusions derived from the inaccurate costs in turn derived from an inaccurate constant and wondering why I'm getting annoyed and don't think you're putting as much thought into this as I am. Okay.
You gotta differentiate between what I'm saying that is my opinion and what I'm saying that is not my opinion. That the statistically best team assuming current players play exactly to their win percentages consists of top players and 1-pointers who do not play is not my opinion. That taking the statistically best approach will make me not-win due to statistical anomalies is unproven, but quite likely. That taking a statistically worse approach will make me not-win is even more likely. What I've been doing is building a model of what scoring looks like based on the new scoring rules, and then applying that model to previous score and game distributions to get a general feel for how this round will play out. When you changed the scoring rules you didn't build this model. And now you keep ignoring or arguing with the model on the basis that it won't win me FPL. That the statistically best model will not win FPL due to the existence of statsitical deviation is a non-sequitur. I can talk in non-sequiturs too: If jaedong gets a game win and an ace match win every game he could score 130 points in a round and end up with a cost of 27. If you didn't buy jaedong you would lose!!1
I don't really want to talk about this anymore, this whole conversation has been somewhat disheartening but I'll wait until Abydos responds before I give up entirely.
|
could you post a list of all players who's cost you adjusted upwards/downwards ?
you adjusted flash from 14 to 10? is that true? if so im definitely picking him up
|
if a player in my antiteam loses a match, do i get points? or does antiteam only affect my points if they win?
|
wow, do a ctrl+f search for flash. he's on just about every team.
if a player in my antiteam loses a match, do i get points? or does antiteam only affect my points if they win?
yes. no
|
On December 04 2009 14:29 lazz wrote:
could you post a list of all players who's cost you adjusted upwards/downwards ?
you adjusted flash from 14 to 10? is that true? if so im definitely picking him up
if they had adjusted the constant down 14.6% from the old inaccurate constant flash would've been 14 points, yeah. with an accurate constant flash is worth 12.08 points under the new scoring based solely on the previous round.
edit: you should note that flash's win percentage was ~92% the previous round, ~22% higher than his lifetime average win %. regression to the mean suggests that at some point he will drop to a 70% winrate or below for a while.
|
|
|
I think Frozean is missing?!
He's playing twice but i can't add him to my team? =(
|
Perhaps Flash is just legitimately ahead of the game now? Skill does progress in this game after all. Clearly you can't keep up a 92% winrate, but there is no reason he has to drop as low as 70%. Just watch a few of his games and you would understand.
Edit: Also, this is progleague and not individual league where the competition is more difficult as you advance leaving you more chance to lose.
|
On December 04 2009 16:25 Vibes wrote:
I think Frozean is missing?!
He's playing twice but i can't add him to my team? =(
Cuteangel on the 2 pointer list.
|
Wow, we're already up to about the same amount of entrants we had last round, with 20 hours and the last-minute rush still left to go.
|
Would you guys rather have a borderline 2300 ELO players (bisu or jaedong) and an underrated top 25 bargain basement player or two cheap 2200+ players (stork and kal)?
|
On December 04 2009 16:59 igotmyown wrote:
Would you guys rather have a borderline 2300 ELO players (bisu or jaedong) and an underrated top 25 bargain basement player or two cheap 2200+ players (stork and kal)?
Pick ace players
stork has competition with jangbi (although jangbi has been sucking) and kal has fierce competition from calm. id go with bisu/jaedong, but remember that both have their costs have been artificially inflated, if bisu/jd perform as badly in r2 as they did in r1 they would only be worth around 6 or 7 points and you wont make any profit
|
On December 04 2009 15:58 integral wrote:Show nested quote +On December 04 2009 14:29 lazz wrote:
could you post a list of all players who's cost you adjusted upwards/downwards ?
you adjusted flash from 14 to 10? is that true? if so im definitely picking him up if they had adjusted the constant down 14.6% from the old inaccurate constant flash would've been 14 points, yeah. with an accurate constant flash is worth 12.08 points under the new scoring based solely on the previous round. edit: you should note that flash's win percentage was ~92% the previous round, ~22% higher than his lifetime average win %. regression to the mean suggests that at some point he will drop to a 70% winrate or below for a while.
hehe that's interesting that you're applying regression to pro starcraft. i dont know how much validity it holds.
also remember that flash didnt play many ace matches last round because KT rolled everyone 3-0 or 3-1.
|
On December 04 2009 16:59 igotmyown wrote:
Would you guys rather have a borderline 2300 ELO players (bisu or jaedong) and an underrated top 25 bargain basement player or two cheap 2200+ players (stork and kal)?
Is "rather" based mathematically or otherwise? If you're going by the math you should do a quick calculation of win rate over a significant sample size of time, bisu's win rate in the 08-09 season was 55-15 or 34-8 excluding winner's league if TLPD is correct. Stork was 33-24 and Kal was 35-24. Kal's win rate isn't dramatically different now, neither is stork's, neither is bisu's, so this seems like a decent way to compare, so let's use it:
Bisu's point differential would be 190. Include ace match appearances (6) and team wins (35) and bisu earned 237 points all year. Not going to do streak points.
Stork's point differential would be 84, Kal's would be 92. Include ace matches (kal 9 stork 6) and team wins (33 for STX and 30 for KHAN) and combined your total is 269 points for both players. Kal probably plays less ace matches now because Calm's a beast and because STX is dumb and sends modesty and hero instead, but whatever.
As long as your dude earns 13.5% of Bisu's points (team wins alone would do it as long as they're not on ACE or KHAN) and win rates are roughly comparable to the previous season, go with bisu and the dude.
edit: or jaedong, he was worth 255 points last year.
|
On December 04 2009 18:04 lazz wrote:Show nested quote +On December 04 2009 16:59 igotmyown wrote:
Would you guys rather have a borderline 2300 ELO players (bisu or jaedong) and an underrated top 25 bargain basement player or two cheap 2200+ players (stork and kal)? Pick ace players stork has competition with jangbi (although jangbi has been sucking) and kal has fierce competition from calm. id go with bisu/jaedong, but remember that both have their costs have been artificially inflated, if bisu/jd perform as badly in r2 as they did in r1 they would only be worth around 6 or 7 points and you wont make any profit
Kal is worth 3.54 points based on an accurate scale. Stork is worth 4.375. Right now they both cost 5. Take from that what you will.
|
Ah nice intelligent reply. The 2 point player is type-b or canata, who performed relatively unimpressively (lack of games?), but statistically I expect them to win at least 50% to high 50%'s of their games.
Shouldn't 34-8 be 116? Stork was 28-16 for 112-32=80. Kal was 28-19 for 74. If those records include ace matches, shouldn't that just inflate the scores by removing loss points?
If they perform the exact same as last season, I'm guessing we'll all perform exactly average, so I'm hypothesizing ELO will be a better predictor than season games and I want to see if it performs above average. I assume that Jaedong will be worth more than 8/9 of Sea or Zero.
Jaedong hasn't been acing, and Bisu's not getting the nod that often, and I'm not comfortable assuming they will always ace. I also personally think that Flash will regress, as hot players tend to, but if I'm going to follow my hypothesis I can't ignore a 2350 ELO.
|
|
|
|
|
|
|
|
|
|
EPT
