EPTOn February 14 2012 04:16 Alejandrisha wrote:
i think you just continued making banshees each patch ^^
i think you just continued making banshees each patch ^^
I rarely makes Banshee in TvP and TvZ.
Win Rates by Race & Length of Game - Page 10
| Forum Index > SC2 General |
|
VPFaith
United States261 Posts
I rarely makes Banshee in TvP and TvZ. | ||
|
p1cKLes
United States342 Posts
The other frustration I have is in TvP, I feel like I am forced into going ghost no matter what unit composition Protoss has to even have a chance in the mid to late game. This doesn’t hold true for Protoss. There’s not 1 particular unit they have to have in order to compete with Terran, but in order to compete with Protoss you have to have ghost. You are not going to beat Protoss in an army vs. army engagement. You might be able to side step it with mineral harass, drops, etc., but if you try to engage in a straight up battle, you are going to lose without the ghost. I don’t feel like this is the case with Protoss. Now that is only my opinion, and I could be wrong, but in almost every game I see or play that goes past the mid game you better get ghost or you’re done. | ||
|
VPFaith
United States261 Posts
On February 14 2012 04:49 p1cKLes wrote: Mid to late game against Protoss and Zerg for that matter frustrate the hell out of me. This graph is very indicative of how I feel. I feel like I have to end the game within the first 15 minutes of the game, because otherwise I’m done no matter how well I macro. It makes me feel like I have to do an all-in 1 or 2 base timing attack. The other frustration I have is in TvP, I feel like I am forced into going ghost no matter what unit composition Protoss has to even have a chance in the mid to late game. This doesn’t hold true for Protoss. There’s not 1 particular unit they have to have in order to compete with Terran, but in order to compete with Protoss you have to have ghost. You are not going to beat Protoss in an army vs. army engagement. You might be able to side step it with mineral harass, drops, etc., but if you try to engage in a straight up battle, you are going to lose without the ghost. I don’t feel like this is the case with Protoss. Now that is only my opinion, and I could be wrong, but in almost every game I see or play that goes past the mid game you better get ghost or you’re done. Well, if any Terran player who is struggling, let me know, I will try to help you out. | ||
|
TheBengalTigger
United States8 Posts
Simply put, a good Terran will often end the game before 20 minutes against Zerg or Protoss. If the game runs to 20 minutes, it is evidence that the Terran is outmatched / not making good strategic choices / whatever. Not saying this is always the case, but food for thought. | ||
|
NeoGeoOdin
Colombia140 Posts
| ||
|
shizaep
Canada2920 Posts
In order for these statistics to be more accurate, you should have some form to represent what percentage of games end at what time interval. I suggest using a scatter plot with each zerg win as a red dot and each protoss win as a green dot. This way, they will still follow a general trend but you will be able to see at what time most games actually end by seeing the largest cluster of dots. I really hope you will update this because I just do not think these are valid statistics at the moment. EDIT: I also see a lot of people saying that T has the weakest late game. I think that they are forgetting that T ends a good amount of their games before 20 mins. Just want to remind everyone to keep in mind that these graphs are not proportionate so the line does not represent win rates in a matchup, just the win rates in x number of games (could be 10, could be 100). | ||
|
tdt
United States3179 Posts
| ||
|
Jarree
Finland1004 Posts
On February 14 2012 11:58 shizaep wrote: The way this is graphed is flawed. For example, you said 100% of ZvPs before 5 minutes result in a win by the zerg. This is misleading because probably something around ~5% of ZvPs end at that stage. (I'm pulling the number out of my head but I know that it is really low and something around like that). I would assume that the vast majority end after 10+ minutes. In order for these statistics to be more accurate, you should have some form to represent what percentage of games end at what time interval. I suggest using a scatter plot with each zerg win as a red dot and each protoss win as a green dot. This way, they will still follow a general trend but you will be able to see at what time most games actually end by seeing the largest cluster of dots. I really hope you will update this because I just do not think these are valid statistics at the moment. EDIT: I also see a lot of people saying that T has the weakest late game. I think that they are forgetting that T ends a good amount of their games before 20 mins. Just want to remind everyone to keep in mind that these graphs are not proportionate so the line does not represent win rates in a matchup, just the win rates in x number of games (could be 10, could be 100). Come on... Click the image to enlarge. There is the game count by matchup/length. | ||
|
Space Invader
Australia291 Posts
| ||
|
smoosh
31 Posts
I could not have said it better myself, Primal666. Statistically, all this data will ever be an indication of is the current metagame. To try to explain this better, particular strategies that people use tend to end the games around the same time. Therefore, when gathering data, game end times start to be an indicator of the strategy people used. Thus, a graph of the most common end times would simply show how long it takes to win with the most common strategies. So if you see terrans with a high win rate at less than 15 minutes, it just means that the strategies most terran are using end the game around 15 minutes. Onto the next thing. On February 11 2012 19:02 [17]Purple wrote: I see the low Terran win rate in the late game as a sign that most of them actually play risky strategies that is intended to kill their opponents at the 10-15 minute mark, but they just last so much longer compared to other races. As an example, a Terran 1-1-1 against a Protoss (I can't site an example right now) would be very effective and can kill of the Protoss at the 14 minute mark if the Protoss doesn't defend properly. However when the contain is broken and Terran retreats, they still stay in the game despite having a very distinct disadvantage of going for a strategy that was intended for an mid game kill but did not work. I might be completely wrong though and if someone can prove/tell me otherwise, please do! This explains the dropoff in win rate for terran quite nicely. First off, if a terran is playing to win the game before 15 minutes, and they fail, they are generally behind. So since the data before 15 minutes indicates the most common terran strategies, data after 15 minutes relates to terran players who are already behind. Also, terrans can turtle better than most races, so when they fail their 15 minutes strategy, it becomes very hard to kill them right away, but easier as the game goes on, explaining the downward slope. I hope this makes sense to people. For shame all of you terrans who used this as an excuse to complain. On a side note, even though this sample size was very small, it actually turned out to be a very accurate indicator of the current metagame (aside from less than 5 minutes). | ||
|
ZeroTalent
United States297 Posts
On February 14 2012 12:11 Jarree wrote: Come on... Click the image to enlarge. There is the game count by matchup/length. Actually, I think showing some sort of 1D scatter plot or histogram showing the time of the wins might be useful... Also playhem people: WHERE CAN I GET THAT GINORMOUS REPLAY PACK. :D :D | ||
|
shizaep
Canada2920 Posts
On February 14 2012 12:11 Jarree wrote: Come on... Click the image to enlarge. There is the game count by matchup/length. You misunderstand what I mean. I see the fact that there are win percentages by race, I'm not retarded. What I'm trying to point out is that it does not indicate at what time the majority of games end. For example, it may be (just pulling random numbers here) that 80% of ZvPs end at 15 minutes. If toss has a 70% win rate in this time frame, it has heavier weight than if zerg has a 100% win rate in the first 5 minutes of the game, where only say 5% of all games end. Read the post, son. | ||
|
ZeroTalent
United States297 Posts
On February 14 2012 13:35 smoosh wrote: This explains the dropoff in win rate for terran quite nicely. First off, if a terran is playing to win the game before 15 minutes, and they fail, they are generally behind. So since the data before 15 minutes indicates the most common terran strategies, data after 15 minutes relates to terran players who are already behind. Also, terrans can turtle better than most races, so when they fail their 15 minutes strategy, it becomes very hard to kill them right away, but easier as the game goes on, explaining the downward slope. I've seen a lot of people make various forms of this argument I'm trying to come up with a way to test this ... I assume I would have to do something tricky like inspect the build order to see if it suggests an all-in or quasi all-in: for instance, building > 8 production facilities before building a third CC. There are other theories I'd like to test (how much difference does APM make? Do the Korean pros play more aggressively? If I get a large enough data set can I account for skill differences somehow?), but (a) I need a bigger data set, and (b) I need to think harder about how to test those theories. | ||
|
ZeroTalent
United States297 Posts
On February 14 2012 13:35 smoosh wrote: This thread needs to be called out, it is statistically invalid and will never hold water no matter how large your sample size or what sort of games you are measuring (ladder, tournament, team leagues, etc.). I want to push back on this idea. Obviously whatever data set we can come up with is imperfect. Starcraft is not played in some sort of experimental vacuum. But nor is are baseball, basketball, or football, and yet people made great strides in making quantitative assessments in player performance. To say "oh, well, the results here just reflect the current metagame" is a cop-out. When Billy Beane built the Moneyball A's in the late '90s (everyone should read the book, not just watch the movie), most teams overvalued speed, defense, batting average, not striking outs and pitcher ERA while undervaluing on base percentage, slugging percentage, and pitcher strikeouts. The only way the A's were able to figure this out was by looking at the numbers. But once the A's started winning, other teams started adopting their player evaluation techniques. In fact, the lack of emphasis on defense promoted by Beane eventually made speed and defense undervalued(!). Does that mean the A's were wrong to look for fat guys who drew walks and hit home runs, simply because those characteristics reflected the "current metagame"? No; Beane's job was to win games in the present, not create an ideal baseball team suited to play in any conditions. The current metagame as it relates to race (dis)advantages will have an obvious effect on the results at any point in time, in the same way that the metagame in baseball affects league-wide statistics over time (steals are up because catcher defense has gotten so bad that good base stealers are valuable; home runs are down because most players who can steal can't hit home runs, and because of the crackdown on steroids; strikeouts are up because every team is now obsessed with strikeouts; etc.). But that doesn't mean we should just wave away what the data show us. If anything, it means we should look more closely at the results. Perhaps there are significant differences in the builds in winning games versus losing games. Perhaps the late game unit composition is different (sadly the SC2 replay format makes it impossible to examine unit composition). But please, please, please, let's let our theorycraft be informed by the results that we have available, rather than just guess as what Ideal Starcraft ought to look like. | ||
|
Ubenn
Canada114 Posts
On February 14 2012 11:58 shizaep wrote: The way this is graphed is flawed. For example, you said 100% of ZvPs before 5 minutes result in a win by the zerg. This is misleading because probably something around ~5% of ZvPs end at that stage. (I'm pulling the number out of my head but I know that it is really low and something around like that). I would assume that the vast majority end after 10+ minutes. In order for these statistics to be more accurate, you should have some form to represent what percentage of games end at what time interval. I suggest using a scatter plot with each zerg win as a red dot and each protoss win as a green dot. This way, they will still follow a general trend but you will be able to see at what time most games actually end by seeing the largest cluster of dots. I really hope you will update this because I just do not think these are valid statistics at the moment. EDIT: I also see a lot of people saying that T has the weakest late game. I think that they are forgetting that T ends a good amount of their games before 20 mins. Just want to remind everyone to keep in mind that these graphs are not proportionate so the line does not represent win rates in a matchup, just the win rates in x number of games (could be 10, could be 100). You say the graphs are flawed but you're also misreading them completely. | ||
|
Sated
England4983 Posts
| ||
|
Bommes
Germany1226 Posts
Every matchup is fine in the early game and there is still a lot to explore in the late game, I don't see any problems. The ghost nerf is a bit saddening though, because I think its very over the top. Also its a bit sad that PvP relies heavy on having a ramp at your base to get interesting, but you can't have everything I guess. edit: Also note that part of the reason why the statistic for terrans is SO bad after 20 minutes is that they often 1 or 2-base allin and in case it gets delayed they will still play it out, although they know they have probably lost. You shouldn't value these statistics too much, they are influenced a lot by current trends of how to play. The thing is, a 1-1-1 is really strong if it hits fast enough. If a Protoss can stall it and slow the push it gets worse and worse, so if you hit 20+ minutes you will lose with a 1-1-1 for sure. That happens quite often. Many ppl value statistics way too much and they DEMAND that something has to change with players not changing their playstyle so the statistics get even. That doesn't even make sense, my mind is blown. Think about how players played 1 year ago and think about how they play now. And there is still so much room to improve and change. I still think that maxed terran vs maxed protoss/zerg might get a problem at some point. Especially Broodlord/Infestor which is just sad to play against. But I don't think its time to change anything right now. Also TvZ early game might STILL be worth to look at, there are really a lot of all ins the terran can do. But I think it is not impossible to solve and usually if zerg loses early they were too greedy in the first place. | ||
|
qwertzi
111 Posts
On February 05 2012 06:22 ZeroTalent wrote: To the Pros, Masters players, and everyone: do the numbers here line up with how you think about the game? Is Terran overmatched in the late game? Does Zerg grow stronger as the game goes on? Should you 6-pool (100% Z win rate in PvZ <5 minute games) or 2-rax every (80% T win rate in ZvT 5-10 minutes) every game? Discuss. your information is misleading. it may very well be that zerg has a 100% winrate vs protoss under 5 minutes, yet, that does not account for failed sixpools.if sixpool fails, the game goes on, at least 5 more minutes ... thus, this statistic says nothing about the success of sixpools. assume only 1 in 10 works, that gives you 10% winrate, of those 10%, 100% are within the first 5 minutes.. nonetheless, interesting stats | ||
|
Marddox
United Kingdom108 Posts
| ||
|
Wegandi
United States2455 Posts
| ||
| ||
