So, the guys over at IEM have released the replays for the tournament pretty much as the last game was finishing.
I've grabbed them and, after fighting with the fact that some of them were made using the Korean client (Blizzard please standardise your replay files!), have some information about the games that may or may not be interesting to you all.
This was done using what I've written over here, which I will be releasing once I've added Starbow functionality and removed the Python dependency.
TL;DR - Pros are consistent - Matchups swing back and forth over time - We can probably find optimal scouting routes - Protoss players hate the middle of the map - People attacked right to left and top to bottom. Most fights happened on the left or bottom of the map.
Item 1 - Pro players are consistent. "Well, duh" you might be thinking, but to me the numbers were striking. Let's look at our favourite build of the semi-finals, the 6+ Gate blink attack. We'll focus on herO's execution for this.
You can see from this that herO's execution of his gateway + blink build doesn't seem to depend on the map too much. 3 games and the Nexus timing is out by at most 2 seconds from each other. It was possible to see from watching the games that he had even the timing of Polt's SCV to avoid the scout! The slowest build was finished at most no more the fastest. From the build overview screen we can see that the whole build is timed to pretty much perfection:
It shows through skill levels, too. Looking at our now eternal friend Scoobers:
What's this down to? Is it lack of stacking? Not paying enough attention? Extra base scouting before depot placement? Nerves?
Switching to Taeja: All within a much narrower window. Wrists be damned.
However, builds aren't everything: Highlighted is Polt vs herO. The W and L I have added to the image to show the builds that won and lost. You can see the builds he won his units were out LATER than when he lost. This suggests that it's not something you can just be better at your build to beat. Which, in honesty, sucks.
Item 2 - There's a time and a place for you.
Zerg seems to have a window where their builds come through. As you can see, as well, even though Zergs didn't feature in RO4 onwards, they won 50% of their games against Protoss. After 20 mintues, though. Zerg won 3/12 games.
Protoss 20 - 13 Terran. A larger difference, here. Terran seem to have more of a midgame presence. If you can live past 10 minutes you stand a chance! It's pretty much 50/50 past then.
Zerg 13 - 8 Terran Interesting for me here is that Zerg lost 0 games before the 18 minute mark.
Also on this build front it looks like certain builds have phased out. The Protoss vs Zerg build of going early 4 gates and attacking with zealots to deny a 3rd was used ONCE during the tournament. It lost. I guess there wre only 5 PvZ games in the whole tournament, though, so don't read much into that!
Item 3 - Watch Towers aren't always the best places to camp out Again, focussing on Blink v T:
So, whether you have your worker on the watch tower or not, it might be worthwhile having a force around there (especially if it's more marauder heavy - research concussive first?) to deny early 2 stalker + probe moveout. It seems that the blink build is only really powerful if there are actually stalkers there, and you can see that there aren't many other pylons around (the others were placed much earlier, maybe different builds?) then it might deny the build before it starts.
I'm actually really interested in this aspect and I will be looking through all the maps over time to see if I can come up with the best scouting routes or locations to either shut down or disrupt particular builds.
Item 4 - Protoss games end up with fights anywhere but the middle of the map Protoss games seem to end up with more straight up aggression or defense. I assume this is because at the start there's a lot of aggressive openings and later on nobody wants to engage a deathball if they can help it:
Item 5 - People like attacking from the right to the left, or from the top to the bottom This one REALLY interests me. Check it out:
More markedly:
Again, pretty strongly on Heavy Rain:
Finally, pray you don't spawn bottom left on Frost:
So those are the questions that could think of to answer with the IEM games.
I'm interested in questions you may have about high level analysis of games. Think about: - Build information - Building Placement - Maps - Player Stats - Anything else you can come up with
Any questions on here I will reply to. If I can I'll add them to this post, too, so they're easily viewable.
Thanks for reading! Post 2 will contain the same for Seat Story Cup!
So, as I was doing IEM's post, SSC released their replays. A look over the game analysis doesn't throw up such striking information as IEM, but there's stuff here we can talk about.
TL;DR - You can always improve your build timings - Game times vs Race Wins is not consistent with the IEM games - That one game on heavy rain broke my analysis - Being aggressive to the left and bottom of the map is consistent with the IEM games
Item 1 - You can always improve your build timings For a case in point here I'm going to look at Harstem's build: Up to and including the Nexus starting at 3:47.
Now, look at herO doing the same opening a few times: This build is 10 seconds faster up and and including the Nexus starting.
What I take away from this is no matter who you are or how well you think you've refined your build, you can always make it faster. Never. Stop. Practicing.
Item 2 - Game times vs Race Wins is not consistent with IEM Game shifts heavily to Terran after early Protoss presense. This is mostly in line with IEM's matches.
Protoss is shut out of the game after 20 minutes, and even struggles after as little as 10. This is completely counter to IEM.
TvZ is neither here nor there for the most part. Zerg have a 5 minute window in the mid-game time-frame. Similar to the IEM findings.
Item 3 - That one game on heavy rain broke my analysis Even so, it looks cool The fight progresses from the edges of the map to the middle over time as bases and units both run out. There's a storm coming, and it's coming towards the middle.
Item 4 - Being aggressive to the left and bottom of the map is consistent with the IEM games It wasn't just IEM where being on the left or bottom spawn meant the fight was at your base more At least there was a fight in the middle here!
Bottom spawn's 3rd takes a beating!
The fight's not all at the bottom left this time, but again left is where the action is
Lastly... Why even bother being at the top of the map? Nothing going on!
That's it from the SSC Replays. Again, if you have anything you want answered, let me know. Try to specify which tournament you're interested in if it's not applicable to both. Thanks for reading my text and looking at my pictures!
This is cool cool stuff. Very interesting as well, I wonder if the attack pattern occurs is due to the camera angle and psychological effect more than anything else
On March 18 2014 08:53 EpicDemente wrote: there is a clear reason why protoss dont like the middle, its normally more open so storms/ff/colossus doesnt work as well
A good point. In a lot of the maps, though, even ones like Habitation Station, there are usually at least some ramps/walls in the middle of the map that can be used. Not quite the same as a gateway + FF wall, though, I'll grant you!
Recently, the creator of the gameheart-map asked for suggestions on reddit for what to do with a feature they are developing that makes it possible to extract replay-information every 5 seconds while a game is being played. I suggested something akin to this - that it might make it possible to compare the game being played with a database of games live in a broadcast. Well, here is a piece of software that looks at replays and try to extract information useful in just this way. Cool!
I think we are way too often in the dark when it comes to builds and execution. Is this a common build? Is it being executed well? Is it generally successful? How do the builds of the two players match up? Is the player performing better or worse than what he usually is? It would bring so much life to the generally uneventful early game of SC2 (as a spectator, that is). And it's pretty damn interesting in itself!
There is so much potential in what statistics about the hundred thousands of games registered in aligulac (or, maybe more realistically, the thousands of games from recent tournaments) could bring to the live casting of SC2. I'm envious of what Dota 2 is able to do with win-rates, timings of items etc. It would take work, but it could enhance the experience of watching SC2 a lot, I think.
With a database of parsed replays, it might even be possible to do some statistical work on the material to extract common/repeating build-orders and thus skip the manual step of defining the parameters yourself. When players break the mold, we would be aware as the database of known builds are unable to recognize timing or build order.
The heatmaps are quite interesting. On a 2 player map, one would generally expect symmetry (even on a 4 player map really, but you'd need more games for that to happen). But if you look at polar night for example, there are quite some differences between the 2 sides. Maybe the sample size just isnt large enough to even out all relevant factors, but it's still quite funny to see.
Yeah, the sample size is small for the heat maps, but the finding is consistent across all games!
My problem with doing all this is that I'm in about Gold League so I don't always know what to ask. The mathematician in me knows how to answer the questions I can ask, but outside of that I'm stuffed!
I'm looking to release the program within a couple of days. I think I've found a way around my issue with py2exe.
On March 18 2014 08:36 Gowerly wrote: I've grabbed them and, after fighting with the fact that some of them were made using the Korean client (Blizzard please standardise your replay files!), have some information about the games that may or may not be interesting to you all.
Can you elaborate on this? I'd like to understand the issue you ran into. I've never noticed any differences between replays generated by different regions.
The output you get when running them through https://github.com/Blizzard/s2protocol to get the events is partially in the language of the client that generated the replay.
If you grab the IEM replays and look at, say, Open Bracket Winner's Round 1 - Funkay vs Leenock, you'll see the replays are taken from a Korean client.
You can check this out without running it through the parser, just look at the replay in SC2 - the map name is in Korean, as are the player's races.
I simply wasn't expecting this, so when doing race analysis I saw there was a reduced number of replays, so it took some time to find out and rename them!
I would be really interested for more research for this:
People like attacking from the right to the left, or from the top to the bottom
I think there was an article about Pro FiFA that says that the player who plays from left to right has an advantage stats wise. It cuold be possible that something like that is the case in sc2 too (fighting from bottom to top could be easier than the other way for example).
On March 18 2014 22:59 Gowerly wrote: You can check this out without running it through the parser, just look at the replay in SC2 - the map name is in Korean, as are the player's races.
Ah okay. If you are willing to pull down the s2ma file you'll find that all Blizzard maps and most major maps have enUS localization files that you could use to automate that translation process. For your work it is probably easier to just map the names by hand though.
Something I noticed is that the "Unit Death"-"heat maps" interpretation in terms of "where do battles take place" is superimposed by the swarmhost/locust attack routes. Especially on the representation for Heavy Rain you can see where the locust rally to die (look at the middle corridor). - Also, just by looking at this image, all the action of the IEM game between Revival and jjakji appears before my eyes. :D
EPT![[image loading]](http://i.imgur.com/jTzq4Rx.png)
![[image loading]](http://i.imgur.com/KI80jGd.png)
![[image loading]](http://i.imgur.com/EDxDgGM.png)
![[image loading]](http://i.imgur.com/p5hNomu.png)
![[image loading]](http://i.imgur.com/oGtr9kQ.png)
![[image loading]](http://i.imgur.com/6WaP5Bd.png)
![[image loading]](http://i.imgur.com/KMK1Dr2.png)
![[image loading]](http://i.imgur.com/MIkKzbz.png)
![[image loading]](http://i.imgur.com/DsTMcU1.png)
![[image loading]](http://i.imgur.com/uJAithq.png)
![[image loading]](http://i.imgur.com/M4vudNe.png)
![[image loading]](http://i.imgur.com/FjDFIDQ.png)
![[image loading]](http://i.imgur.com/pWxXsb3.png)
![[image loading]](http://i.imgur.com/rvrAIJj.png)
![[image loading]](http://i.imgur.com/QTq8WMY.png)
![[image loading]](http://i.imgur.com/Z9PWwEE.png)
![[image loading]](http://i.imgur.com/RdrOY4H.png)
![[image loading]](http://i.imgur.com/ICjT1bG.png)
![[image loading]](http://i.imgur.com/buTp8Tq.png)
![[image loading]](http://i.imgur.com/1av6SXJ.jpg)
![[image loading]](http://i.imgur.com/qh6CM4a.png)
![[image loading]](http://i.imgur.com/O1WIzSR.png)
![[image loading]](http://i.imgur.com/Qm2Q5kK.png)
![[image loading]](http://i.imgur.com/zUeQSgT.png)
