|
SpawningTool holds the largest collection of StarCraft 2 pro-level replays. The site uses machine learning and the information stored in the replays to generate statistics, label build orders, and organize games by series/event. A new feature called Build Orders allows anyone to create and share their custom build orders!
Over the past months, we have been providing stats for the Team Gravity Fight Night showmatches for the casters to use during the event. With the success of that partnership with Team Gravity, we decided to collect some statistics for the RO8 in the Global Finals this Friday!
HerO vs Classic
1. herO is a fan of blink stalkers, going for them in 48% of his games. (source)
2.In recent matches vs classic, herO has not won a game that goes over 20 minutes. (source)
3. In recent matches vs classic, herO has a higher winrate (66%) in games from 8-12 minutes long. (source)
4. Classic has a very solid playstyle that does not have many weaknesses (overall 69.% winrate vs pro-protoss players), but has a 28% winrate in games from 12-16 minutes long. (source)
5. Classic enjoys opening with a stargate in 50% of his games (oracle 39%, phoenix 11%). (source)
Innovation vs Life
1. Innovation sure loves his proxy 2 rax, going for it in 45% of recent games vs zerg. (source)
2. Innovation recently has been doing a mine drop in tvz. He has been doing these in 27% of his games. (source)
3. In his 11 most recent tvz replays, Innovation has a 91% winrate, losing only one game to soO at the intel extreme masters.
47% of games Innovation and Life play range from 20-25 minutes long. (source) (source)
4. Innovation has won every game (in our database) under 8 minutes vs life, which were both proxy barracks play. (source)
Rain vs sOs
1. sOs, even though being a cheesy guy, is very strong in the late game, winning all games (in our database) 20 minutes and longer, including one versus herO, which he might be facing later in the bracket. (source)
2. Rain has a higher winrate vs sOs in the midgame(16-20 min (60%)) than in the early game (0-16 min). (source)
Hydra vs Rogue
1. Hydra likes to play on the safe side, going pool first 52% of his games. (source)
2. Hydra is phenomenal at defending allins and pressure, winning 71% of games 8-16 minutes long. (source)
3. Rogue is a very risk oriented player. He does some type of zergling rush in around 20% of his games and goes for pool first in only 38% of his games. This means that if he goes pool first there is a large chance (about 50% chance) that he will be on the aggressive front that game. (source)
4. Even though Rogue cheeses, his end game skills are also hard to match. In games 20-25 minutes, he is 6-0 in games vs high level koreans such as hyun and soO. (source)
Thanks for reading! We always appreciate feedback and would like to know if you guys are interested in seeing stats for the Ro4 and finals 
|
Quality post! Yes please, more stats I love it!
|
Interesting. Would love to see the stats for Ro 4 and Grand Finals.
|
|
|
I need like a billion more StarCraft stats in my life. This is a good start. Yeesss... a good start.
+ Show Spoiler [Website feedback] +
Your website seems like an amazing resource that I'll be referring to a lot in the future, but it can be difficult right now to tell which BO belongs to which player. For instance, on the Rogue vs Hydra overview page, Rogue is sometimes listed first in the replay title, but Hydra is always listed first in the player list, so I'm not sure in what order the builds are being presented in.
Or the herO vs Classic page, there's a game that lists "Blink Stalkers/Dark Templar Rush/Zealot Archon/Phoenix Opening." Not being a connoisseur of PvP, that seems wide open to interpretation in terms of who opened with what and transitioned into what.
If you could have something like two separate lines per game, and everything on the first line is what Player A goes, and everything on the second is what Player B goes, that would make it super clear at all times.
|
|
|
On November 05 2015 16:01 pure.Wasted wrote:
I need like a billion more StarCraft stats in my life. This is a good start. Yeesss... a good start.
Your website seems like an amazing resource that I'll be referring to a lot in the future, but it can be difficult right now to tell which BO belongs to which player. For instance, on the Rogue vs Hydra overview page, Rogue is sometimes listed first in the replay title, but Hydra is always listed first in the player list, so I'm not sure in what order the builds are being presented in.
Or the herO vs Classic page, there's a game that lists "Blink Stalkers/Dark Templar Rush/Zealot Archon/Phoenix Opening." Not being a connoisseur of PvP, that seems wide open to interpretation in terms of who opened with what and transitioned into what.
If you could have something like two separate lines per game, and everything on the first line is what Player A goes, and everything on the second is what Player B goes, that would make it super clear at all times.
Yeah, we have actually been working on that a bit as of late. That whole page needs a bit of tweaking to make it a lot easier to use. Thanks for the advice!
|
Nice read! Thanks for posting
|
Czech Republic12129 Posts
Cool, and I thought the day cannot be better  Well written, well done, thanks!
|
|
|
Awesome ! Thanks for that work, keeps the hype strong !
|
Awesome stats! Really fun to read ahead of the games, so hyped for the games and how Life will defend against inno's two rax ^^
|
United Kingdom36161 Posts
love little stats like this
|
Very neat statistics!
But how do you end up with 91% win percent when the sample size is 10 games? ;D Perhaps one opponent got stomped so badly it was 110% victory for INno? The correct answer is that it was really the last 11 games.
|
Cool job guys, keep it up
|
Czech Republic12129 Posts
On November 05 2015 21:59 sagi wrote:
Very neat statistics!
But how do you end up with 91% win percent when the sample size is 10 games? ;D Perhaps one opponent got stomped so badly it was 110% victory for INno? The correct answer is that it was really the last 11 games.
Rounding issue
|
On November 05 2015 21:59 sagi wrote:
Very neat statistics!
But how do you end up with 91% win percent when the sample size is 10 games? ;D Perhaps one opponent got stomped so badly it was 110% victory for INno? The correct answer is that it was really the last 11 games.
Yeah, there were 11 games in total. 10 of them were wins.
|
On November 06 2015 01:57 deacon.frost wrote:Show nested quote +On November 05 2015 21:59 sagi wrote:
Very neat statistics!
But how do you end up with 91% win percent when the sample size is 10 games? ;D Perhaps one opponent got stomped so badly it was 110% victory for INno? The correct answer is that it was really the last 11 games. Rounding issue
I updated it, It should be fixed now :D
|
Siiiiick, keep up the good work!
|
Czech Republic12129 Posts
On November 06 2015 04:55 Morn_sc wrote:Show nested quote +On November 06 2015 01:57 deacon.frost wrote:On November 05 2015 21:59 sagi wrote:
Very neat statistics!
But how do you end up with 91% win percent when the sample size is 10 games? ;D Perhaps one opponent got stomped so badly it was 110% victory for INno? The correct answer is that it was really the last 11 games. Rounding issue I updated it, It should be fixed now :D
Good job! =)
|
Would be awesome if the casters at Blizzcon used this info :D
|
"The site uses machine learning and the information stored in the replays to generate statistics, label build orders, and organize games by series/event."
Where does the machine learning fit in? The only thing I can think of is if the labeling of builds as "cannon rush", for example, is autonomous.
|
On November 06 2015 23:38 Hwyk wrote:
"The site uses machine learning and the information stored in the replays to generate statistics, label build orders, and organize games by series/event."
Where does the machine learning fit in? The only thing I can think of is if the labeling of builds as "cannon rush", for example, is autonomous.
You're right! We are using machine learning to try to label some types of build orders in the system. We have used a mix of rule-based systems, manual entry, and machine learning to get coverage over build order labels. I don't have hard stats to back this up, but in general, we found that rules work better for labeling build orders, and our classifier seems to work best with compositions.
|
|
|
Very interesting stuff. Do keep posting similar stats!
|
|
|
|
|
|
|
|
EPT![[image loading]](http://imgur.com/NQ7lLwN.jpg)
