Smurfs have been around since forever, and have a lot of culture accompanying them. Remember the ICCUP Who is Who? threads--curated by the penetrating knowledge of the savant roMAD?
Well, I hope to bring some of that back--and maybe more. I've begun developing a tool called vroMAD (virtual roMAD), which extracts hotkey information from replays and uses this to calculate the similarity between the hotkey setups of players in different replays. This tool relies heavily on the sc2reader library, so I owe ShadesofGray and the sc2reader contributors many thanks.
How does it work? First, to analyze a replay with unknown players, vroMAD needs a repository or folder containing reference replays containing players with known identities. It extracts hotkey information using sc2reader, and quantifies the hotkey setup of each player in every replay. It then uses the information "learned" from each of the players in the reference replays to compute a similarity score against the players in the unknown replay. Note that each reference replay contributes at least two players to the reference set. So, for say around 100 reference replays, there will be around 200 reference players to be compared against.
Update 2013/09/23 I'm currently in the process of migrating computer setups, and my internet access has been spotty for the last couple of days. I hope to have everything sorted out by the end of the week so that I can continue building binaries on my Windows machine.
If you wish to run the Python source directly, the latest development version is on github. https://github.com/eqy/vroMAD
There is a Python2 branch if you only have Python2. This branch will run fine in the Python2 interpreter, but won't compile to a distributable binary.
See the following video for a quick 1-minute demo of how it's used.
Note that I'm testing it with known replays to highlight how the similarity measure works. Also note that your loading times will vary: it will probably take around two minutes to load 100 replays.
The most important thing to suggest a player's identity is the first few entries in the similarity ranking: if the ranking is dominated by a single name in the first few spots AND the score is quite high (> 0.96/0.97), there's a good chance you've found the player
There are still issues with some replays recently uncovered that seem to be malformed. There are a couple of these that will crash sc2reader+vroMAD from the Group 1 and Group 2 folder of the Dreamhack summer pack. There's a battle.net forum thread about this: http://us.battle.net/sc2/en/forum/topic/10014611448
If you encounter any errors, please respond with either a post in this thread, going into as much detail as possible about the exception so that I may be able to replicate it. Malformed replays, as mentioned earlier, should trigger an exception pop-up. If that's the case, a screenshot of the popup should give enough information.
Github issues are also acceptable.
If you have a replay that crashes the program, sharing it would be great, but I understand that some replays are sensitive (this is why this is a desktop application and not a binary, so professionals don't have to distribute replays they don't want to).
If you have any other questions, feel free to ask!
I remember the Who is Who thread. And i have been longing for a similar thread to pop up. And how awesome isn't it that this time it is automated! Very nicely done!
I want to test the program, but it seems it isn't compatible with 32 bit windows?
On September 20 2013 08:21 Megashira wrote: I remember the Who is Who thread. And i have been longing for a similar thread to pop up. And how awesome isn't it that this time it is automated! Very nicely done!
I want to test the program, but it seems it isn't compatible with 32 bit windows?
Right now the build is x86_64, I'll see if I can have a 32-bit version up soon.
I have nothing else to say about the program itself that hasn't already been said, so as far as it goes, it's neat, yes.
Have you considered, however, should this program prove very accurate, that it may have negative effects for the pro scene as a whole? I believe that everyone has a right to practice anonymously, even on the ladder.
On September 20 2013 23:57 shadymmj wrote: I have nothing else to say about the program itself that hasn't already been said, so as far as it goes, it's neat, yes.
Have you considered, however, should this program prove very accurate, that it may have negative effects for the pro scene as a whole? I believe that everyone has a right to practice anonymously, even on the ladder.
On the other hand, it can easily catch people maphacking while smurfing. Looking at the preview of Jaedong, it doesn't give you 100% identification, just probability, which is good.
On September 21 2013 00:51 mechengineer123 wrote: I wish people would just let pro's practice anonymously.
I'm sure the pros already know who most of the different barcodes at the top of the ladder are. They run in to eachother and practise together enough that it's not possible to keep it sevret from eachother. This is just for curious TLers who love to speculate and discuss SC2.
With WCS now releasing replays and other tournaments like Dreamhack periodically releasing theirs, there should be a ton of good replays with identified players to be used in this program. I'm really interested to see what comes up from this.
On September 21 2013 00:51 mechengineer123 wrote: I wish people would just let pro's practice anonymously.
I'm sure the pros already know who most of the different barcodes at the top of the ladder are. They run in to eachother and practise together enough that it's not possible to keep it sevret from eachother. This is just for curious TLers who love to speculate and discuss SC2.
Watch last Meta.
On September 21 2013 01:52 Brian333 wrote: What are those gun-like sensor things they use at stores to read barcodes called? That should be the name of this program.
On September 21 2013 00:51 mechengineer123 wrote: I wish people would just let pro's practice anonymously.
There's no real way to practice anonymously unless you're talking about in-house team games, which again, isn't very anonymous.
Realistically the only people who are going to utilize this to any value are other pros anyway. I think it's important, especially for foreign pros, to be able to say "Ya know, I haven't really had any tournaments I could excel in, but I've beaten IM Yoda 3 times this week when previously I'd never beaten him." It's another way to show improvement, and I'm all for more legitimate ways to show improvement.
I think the barcode/smurf nonsense is ridiculous, personally. Especially for the top of the top. They have plenty of available practice partners already at the top that they don't have to ladder for anything but mechanical practice. There's no reason they need to be anonymous to practice mechanics. I'm sure it might be useful, but it's simply not necessary and they're not hurt when someone knows who they are.
EPT![[image loading]](http://i.imgur.com/dIa5Rlp.png)
![[image loading]](http://upload.wikimedia.org/wikipedia/commons/1/13/Barcode-scanner.jpg)
