Actual FTA Data

[Sakiehl]

(10-03-2012 07:13 AM)Ravernoth Wrote:  

(10-03-2012 06:30 AM)Sakiehl Wrote:  Why is there so much skepticism here?

Just because it doesn't match the forum's consensus opinion doesn't mean the data isn't valid.

I've always thought that the FTA problem was overstated, so, naturally, the data strikes me as accurate

Hey Sakiehl,

Look at your matches in the World Cup:

3 P1 games, 3 wins
3 P2 games, 1 win (which took 90 turns compared to 33 turns to win as P1 against the same player)

Did you really not feel the extent of FTA when playing the same player simultaneously as P1 and P2?

That's a fair point, I'm not sure why I was in a contrary mood this afternoon.

Mostly, FTA just hasn't struck me as a big deal. 60% isn't a negligible amount and I'm excited to see what they come up with to counteract it.

I am also curious to see how the numbers have changed in more recent weeks. My impression is that people have gotten better at early, smarter rushes (not the all out rushes right away).

[GreatGonzales]

(10-03-2012 12:18 PM)CombatEX Wrote:  

(10-03-2012 09:41 AM)Necrocat219 Wrote:  Just throwing it out there, considering that for all players it would have taken from their first 30-100 games to fit into their correct league, there will have been massive win and loss streaks that impact this data. When playing players of higher or lower skill level you were already pretty much guaranteed a win or a loss regardless of going first or second.

That's a great point actually. Think of how many thousands of games have a 50:50 P1 : P2 ratio simply because the players won 100% of their games due to skill discrepancy. Players that go on 50-2 streaks or even longer (many players when trying to get to their correct placement) will skew the results towards 50:50 since they win regardless of P1 or P2 assignment.

Yeah, wow, didn't think of that. I estimate that this would have a large impact on the data if not controlled for. Alex, is it possible for you to control for players who have settled into their league? Perhaps the way to do it is to exclude all matches where players exceed a certain rank differential from the calculation.

[Necrocat219]

(10-03-2012 03:16 PM)GreatGonzales Wrote:  

(10-03-2012 12:18 PM)CombatEX Wrote:  

(10-03-2012 09:41 AM)Necrocat219 Wrote:  Just throwing it out there, considering that for all players it would have taken from their first 30-100 games to fit into their correct league, there will have been massive win and loss streaks that impact this data. When playing players of higher or lower skill level you were already pretty much guaranteed a win or a loss regardless of going first or second.

That's a great point actually. Think of how many thousands of games have a 50:50 P1 : P2 ratio simply because the players won 100% of their games due to skill discrepancy. Players that go on 50-2 streaks or even longer (many players when trying to get to their correct placement) will skew the results towards 50:50 since they win regardless of P1 or P2 assignment.

Yeah, wow, didn't think of that. I estimate that this would have a large impact on the data if not controlled for. Alex, is it possible for you to control for players who have settled into their league? Perhaps the way to do it is to exclude all matches where players exceed a certain rank differential from the calculation.

I think that they already are aware of what I said. statistics since the last patch will give realistic fta values

[Harti]

(10-03-2012 06:30 AM)Sakiehl Wrote:  Why is there so much skepticism here?

I wasn't being skeptical :o

It is just I happen to work with very similar massive data analyses these days where I'd have to remove bots and other noise first. I figured there is some analogy and ported it to this discussion, trying to be helpful

(10-03-2012 05:11 AM)onealexleft Wrote:  And if you're proposition is correct? Doesn't really change the gifted+ data. You aren't exactly a new player at this point if you're master+ either.

If I recall correctly, Master+ Players can and will (once in a while) get paired to Placement Matches as well. At least I've played against some of these.

But I fully agree with you, this should not affect much in the higher leagues.


Another possibility is that people maybe don't know how to abuse FTA. They're busy with their own game and stuff... so yeah, they probably won't all be trying to pick off the predeployed Medics to get an edge in a game.
I've entered so many games as P2 and only very rarely I could spot a Runner threat my Medic/Sniper on turn 1.

[Ravernoth]

(10-03-2012 03:16 PM)GreatGonzales Wrote:  

(10-03-2012 12:18 PM)CombatEX Wrote:  

(10-03-2012 09:41 AM)Necrocat219 Wrote:  Just throwing it out there, considering that for all players it would have taken from their first 30-100 games to fit into their correct league, there will have been massive win and loss streaks that impact this data. When playing players of higher or lower skill level you were already pretty much guaranteed a win or a loss regardless of going first or second.

That's a great point actually. Think of how many thousands of games have a 50:50 P1 : P2 ratio simply because the players won 100% of their games due to skill discrepancy. Players that go on 50-2 streaks or even longer (many players when trying to get to their correct placement) will skew the results towards 50:50 since they win regardless of P1 or P2 assignment.

Yeah, wow, didn't think of that. I estimate that this would have a large impact on the data if not controlled for. Alex, is it possible for you to control for players who have settled into their league? Perhaps the way to do it is to exclude all matches where players exceed a certain rank differential from the calculation.

Necrocat, I think you've hit on what the rest of us were missing.

The long win streaks and 90%+ records of players slowly rising up the leagues will really skew the FTA figures to 50:50.

Even looking at games since the patch wouldn't remove this problem since there are many players still ranked below their actual skill level and winning most of the time. However it should give a significantly more accurate FTA figure.

Perhaps Fluffy league shows 65% FTA because there are less players in Fluffy with the skill differential to skew the figures?

A question for Alex and Adam: When running the stats for Super Titan league, does it look at:
A) All the games of everyone currently in ST league
or
B) Only ST vs ST games

A) would include older games vs non-STs while rising up the ranks and would be inaccurate.

The best metric would be to only look at:

- Games where both players' skill ratings are close, AND have stabilized.

But since that is probably too complicated, a good approximation to reduce skill mismatches would be to run the figures for:

- Only ST vs ST games since the patch

[Demon]

(10-03-2012 09:41 AM)Necrocat219 Wrote:  I do really appreciate the release of the data and it's really useful, and to be frank I've been really enjoying the solution we've already been testing.

Just throwing it out there, considering that for all players it would have taken from their first 30-100 games to fit into their correct league, there will have been massive win and loss streaks that impact this data. When playing players of higher or lower skill level you were already pretty much guaranteed a win or a loss regardless of going first or second.

^This.

I'm something like 22-1 in the past 3 weeks or so in 1v1 matches, even against top 10 Gifteds (I'm in Clever) regardless of turn order.

[Xpander]

People have been discussing FTA for chess for over a century and the jury is still out on that one...probably until the game gets solved like checkers. I think it's a bit premature to make any conclusions, at the very least until we see some more targeted metrics. More specifically, i'd be interested to see data after patch targeting players who consistently rank in the upper top 100.

[Necrocat219]

(10-07-2012 08:36 AM)Xpander Wrote:  People have been discussing FTA for chess for over a century and the jury is still out on that one...probably until the game gets solved like checkers. I think it's a bit premature to make any conclusions, at the very least until we see some more targeted metrics. More specifically, i'd be interested to see data after patch targeting players who consistently rank in the upper top 100.

The only reason for chess having a discussion on whether FTA exists is because of how different people think. There is FTA in chess. Fullstop. You can see exactly what moves both players do, and white has the tempo. White finishes setting up first. White gets first attack off unless black tries to surprise them (which as you can see each other is difficult) Whilst white has to outplay black once to gain a significant advantage, black has to outplay white twice to do so.

The solution that chess has to FTA is draws; I can't remember where I read this but I think the drawrate for grandmaster level chess is about 30%. This is because the games are so close. now the win % for white is only a few above black, but if no draws existed it would be higher.

[bmike]

I would say - let's all stop guessing at this point. It's clear the dev team is a) aware that this is an issue b) going to make some changes.

With B) whatever we speculate will be old news when the new changes arrive. With A) you can be sure they will re-evaluate things once the adjusted game play has 10,000 games complete and can bin the data for players that have played more than 50 games and players that have played more than the mean number of games. FTA matters to the players that play the most games - the new players shouldn't be studied so closely or perhaps discarded entirely since the advantage of knowing the rules and basics of strategy are far more governing than who goes first for all but a sliver of the games a player with less than 40 wins will show.

[Necrocat219]

Oh yeah if it counts for anything, I used my total wins, the 60% that are said to be the FTA ratio since the start and when I think I got to my skill level to estimate what my personal FTA might be...

total wins - 165
approximate wins before finding balanced place - 90-110
60% of 165 = 99
100/2 = 50 wins
99/2 ~ 50 wins
so my estimated FTA % is 50/65 = 77%

Of course I have to do upper and lower bounds because I'm sad and this is an estimate that I'm making for no good reason because they are going to release the actual FTA value anyway:

90/2=45 wins
99-45~55
55/75=73%
110/2=55 wins
99-55~45
45/55=81%

So my estimate for the record is that my FTA lies between 73% and 81%. I'm in Masters.

---

@bmike: Sorry but I'm bored and I know doing this was sad but I felt like it @^@