|
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!
|
Great work!
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.
Nice work 
|
|
|
How did I miss that for 16hours????
Can I book a night with your program? :O
|
United States23455 Posts
Wow this is sick! Nice work.
|
nice stuff.
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
|
Yeah, the sample size is small for the heat maps, but the finding is consistent across all games!
How many sample per maps do you have? I'm not a statistician, but from what I learn of it you can't take any meaningful assumptions under 30 samples.
I'd be interrested to see it over more games/tournaments or even some GM and Master ladders.
|
What if GGTracker implimented a public information pool simular to this?
Like all replays uploaded would be anaylzed, and players could click league / map and look at the same data for any uploaded material....
I mean it may not be as indepth or as cool as yours, but anything simular would be so freaking awesome :D
|
Yeah, the sample size was all 126 games from the IEM weekend, which of course gives not that many games per map.
|
|
|
|
|
|
EPT
