|
On February 05 2012 08:44 teddyoojo wrote:Show nested quote +On February 05 2012 08:37 VenerableSpace wrote:On February 05 2012 08:07 Netsky wrote:
Nice
Your analysis shows that Terran cannot play late game. Wish Blizzard would address this. It also shows how dominant Terran is in anything but the late game. two sides to the coin by your logic. well i see it as: allins have a way too high risk/reward factor. in every race. terran is at a disadvantage lategame so they allin/timingpush way more and often dont dare to even think about a third base.
What..? Of course they'll have a 3rd base and probably even a 4th base by 20 minutes..
|
Do this with the playhem data! I don't think the MLG data is very representative of the metagame, considering most koreans sent are terrans (so thus terran is the most skillfully represented race at MLG).
The TvP curve makes sense overall but I think T's midgame advantage is more like 55-60% while P's lategame advantage is more like 75-80%. At the same time I think Z's lategame advantage against T is much less than what that curve shows, perhaps more like 55-60%.
|
Something to keep in mind is that terran has very few methods of active scouting. You can scout your opponent for a given instant with scan but there is no map control scout like overlords and observers for terran. This leads to terrans favoring aggressive builds so that we can poke at our opponents and react to what we see. This is probably one of the reasons terran late game suffers because it is really hard to play passively with terran. Late game terran armies can go toe to toe with zerg armies so I think it is mainly a play style thing than a balance thing.
|
On February 05 2012 07:50 Gyro_SC2 wrote:
With sc2gear can we know the pourcentage of win of a strategy? Can you know the unit composition at the end of each game?
I could get the win percentage for a given build order in a given matchup. The default is to look at the first five buildings, but you can tell it to look for more. But the sample size issues there would be even worse; it might be useful on the Playhem sample set.
Unit composition is impossible; the replay file doesn't contain information on when units die or are cancelled. We'd have to have access to the in-game engine modeling the game, or something similar to the old Brood War API.
|
this is pretty much excatly what i as a terran master would expect.
i checked 800 games with sc2gears:
5-10 min 66%
10-15 min 61%
15-20 min 64%
20+ min 42%
|
Zerg dominating the macro game indeed
|
On February 05 2012 08:19 magnaflow wrote:
Is it possible to do this with all the playhem replays? 35k is a substantial number
I feel while a larger sample size is needed, you can't really include half these daily games.
You run into situations of bronze vs grandmasters, diamonds vs platinums, etc.
I would like to see strictly GSL/GSTL stats. These would reflect stats of the best players in the world playing.
|
I dont find these enharantly useful yet.
Specifically i would be interested in:
What is the P winrate in a 5 minute game, 10 minute game, 15 minute game etc. vs Zerg or Terran.
The fact that terran is eating shit on that graph in the late game doesnt identifiy if they lose more to Z late or P late.
Edit: nvm, thanks oracle
http://i.imgur.com/8Wfcc.png
|
On February 05 2012 10:00 Greenei wrote:
this is pretty much excatly what i as a terran master would expect.
i checked 800 games with sc2gears:
5-10 min 66%
10-15 min 61%
15-20 min 64%
20+ min 42%
T op early game? o.O
it will be very hard to draw solid conclusino from stuff like this. for example, if every terran goes 1-1-1, they will have a huge winrate in the 10muintue range, but if thye fail a 1-1-1 but stay in the game for another 10 minutes, failing a rush like that should probably not reflect in good lategame win rates.
|
Regarding Terran, I think it's very hard to play late game compared to other races, especially TvP late game is very very hard, but there should be no major imbalance.
|
On February 05 2012 10:23 Roxy wrote:
I dont find these enharantly useful yet.
Specifically i would be interested in:
What is the P winrate in a 5 minute game, 10 minute game, 15 minute game etc. vs Zerg or Terran.
The fact that terran is eating shit on that graph in the late game doesnt identifiy if they lose more to Z late or P late.
Was about to bitch at you, then realized you likely didn't click the image. If you click the image, a graph shows up that does show the PvT, etc winrates at X time.
|
I'm glad people like this stuff!
I realize there's going to be issues with skill gaps in a large dataset like playhem or even MLG, but it's just not possible to get a large enough sample of replays any other way. TLPD has 2000 games/month, that's about 2x the number of games in MLG Providence. For differences in skill to alter the win rates, it would have to be the case that either (a) top players consistently play differently (e.g. more aggressive) if they know they are heavily favored in early rounds, or (b) the top players in playhem/MLG had different race composition from the Dead Money. I'm not sure either of those is very likely.
|
On February 05 2012 10:24 Roxy wrote:Show nested quote +On February 05 2012 10:00 Greenei wrote:
this is pretty much excatly what i as a terran master would expect.
i checked 800 games with sc2gears:
5-10 min 66%
10-15 min 61%
15-20 min 64%
20+ min 42% T op early game? o.O it will be very hard to draw solid conclusino from stuff like this. for example, if every terran goes 1-1-1, they will have a huge winrate in the 10muintue range, but if thye fail a 1-1-1 but stay in the game for another 10 minutes, failing a rush like that should probably not reflect in good lategame win rates.
It's not about succeding a rush and winning or failing and losing... Most "rush" or early timings are not designed to give you the complete win but more like giving you an edge late game - at least at high enough lvl play, which is the case for MLG. All in all, if we strictly follow those curve, T is strong early, weak late.
|
On February 05 2012 10:26 Orracle wrote:Show nested quote +On February 05 2012 10:23 Roxy wrote:
I dont find these enharantly useful yet.
Specifically i would be interested in:
What is the P winrate in a 5 minute game, 10 minute game, 15 minute game etc. vs Zerg or Terran.
The fact that terran is eating shit on that graph in the late game doesnt identifiy if they lose more to Z late or P late. Was about to bitch at you, then realized you likely didn't click the image. If you click the image, a graph shows up that does show the PvT, etc winrates at X time.
thanks for letting me know
i guess now my complaint about lack of information is that, for example:
you could have 10 games ending at 10 minutes with terran winning 7 of the 10 (70%), then 100 games end at 20 minutes with zerg winning 65 (65%) and the graphs would still look to be in favor of terran.
actual cumulative winrates are in favor of Z winning 61% of the time in this example.
perhaps this is just me. i fiend for sc2 content.
im so stoked about all of the statistical information coming out lately.
|
On February 05 2012 10:28 WhiteDog wrote:Show nested quote +On February 05 2012 10:24 Roxy wrote:On February 05 2012 10:00 Greenei wrote:
this is pretty much excatly what i as a terran master would expect.
i checked 800 games with sc2gears:
5-10 min 66%
10-15 min 61%
15-20 min 64%
20+ min 42% T op early game? o.O it will be very hard to draw solid conclusino from stuff like this. for example, if every terran goes 1-1-1, they will have a huge winrate in the 10muintue range, but if thye fail a 1-1-1 but stay in the game for another 10 minutes, failing a rush like that should probably not reflect in good lategame win rates. It's not about succeding a rush and winning or failing and losing... Most "rush" or early timings are not designed to give you the complete win but more like giving you an edge late game - at least at high enough lvl play, which is the case for MLG. All in all, if we strictly follow those curve, T is strong early, weak late.
sometimes i dont express my sentiments correctly.
i just meant that the charts cant account for the strategies and metagame at the moment, and the fact that many people of a particular race may be losing the late game may very well be the result of the decisions they make in the early game that compromise the late game (not that the particular race is actually weak in the late game).
for example, if someone went for a canon rush every single game. failed it. played for another 15 mintues and then lost, it wouldnt be accurate to conclude that protoss is weak in the late game.
|
I think a Win rate-game length graph would be useful. If nothing else, we can see how big a disparity is with mid game and late game TvP, as people say. Though I guess the first graph is somewhat telling of that. Seems like Terrans just have a harder time the longer the game goes.
|
On February 05 2012 10:18 Orracle wrote:Show nested quote +On February 05 2012 08:19 magnaflow wrote:
Is it possible to do this with all the playhem replays? 35k is a substantial number I feel while a larger sample size is needed, you can't really include half these daily games. You run into situations of bronze vs grandmasters, diamonds vs platinums, etc. I would like to see strictly GSL/GSTL stats. These would reflect stats of the best players in the world playing.
Then we wouldn't have a big enough sample size and one race winning over another could skew results in favor of that race just by one game alone. I don't care if we see bronze vs grandmasters, I think the larger the sample size the better. Of course out of those 31.5k games it would be nice to see the Top 8 only analyzed, or even the just the 2nd and 3rd rounders and above because most 1st rounders who lose will be very bad players.
|
On February 05 2012 10:25 aTnClouD wrote:
Regarding Terran, I think it's very hard to play late game compared to other races, especially TvP late game is very very hard, but there should be no major imbalance.
i agree with this.
terran seems just insanely hard to play lategame...
i think because terran doesn't really have anything that's good in small numbers. like a single infestor or HT can single handedly shut down a drop, or at least quickly deal significant damage to a group of units... but terran doesn't have anything capable of this.... except seeker missile... but raven is just too vulnerable to be a viable unit.
ergo you're forced to using MMM, and it feels like you need a critical mass of MMM to kill 3/3 zealots etc.
|
Haha interesting read. Zerg 6 pools 
As a terran, i find late game extremely difficult due to slow reinforcements. I find that the only way i win, is if i have dealt damage early on and i am just outmacro-ing my opponent OR if my opponent makes a mistake during a deathball engagement.
|
those graphs are not representative
and thats why:
terran is the hardest race to kill
eventhough the game's alrdy "over" it often takes 10 more mins to finish off a terran player. u need to wait for hightech to kill him. just watch some stream of pro zergs playing against much lower opponents on ladder, alot of times the game is alrdy "over" at 10-15min mark,cuz they cleaned up their pushes so hard that they never come back. but the games goes on untill broodlords or other stuff is out.
on the other side terran allins are very powerfull and they mostly hit instantly and force a gg immediatly
lets take a 11/11rax: 11/11 wins u the game -> +1 for 5-10min win, but if 11/11 rax fails hard its not counted as 5-10min lose, its counted as 10+ mins lose. thats just not very representative at all. also u have to mention that its a tournament and there alot of players fighting for their tournament life.
lets take 1-1-1
push wins at 10min -> +1 to 10-15
push loses at 10min -> stay in game 10more mins and get +1 lose at 20min
i still think that it shows abit a small tendency but its by far not that much like 60% 15-20min and just 40% 20+, its just because 20+mins contains all the turtle terrans trying to come back.
tldr:
terran wins earlygame -> +1 for 5-10mins -> high win % earlygame
terran "loses" earllygame -> +1 for 10+ mins -> higher lose % lategame
|
|
|
|
|
|
EPT
