StRyKeR's Blog

I haven't streamed in a while because of my internet connection. It should be okay now so I might stream a little bit more. It's definitely harder to find time to play now that I have a full-time job!

Anyway, here are some of my recent ZvT replays. Games are C+/B- level. There are 20 in the pack, mostly wins.

http://dl.dropbox.com/u/4422/TL%20blog/11072009.rar

Recommendations
#1

I know Day[9] advocates going mass muta + hydraling against Terran mech and not trying to be "cute" with defilers and lurkers. I agree with his general point, but if the Terran mechs, somehow gets 3 base, and starts camping like a boy scout, you have to start playing "cute".

Some people advocate attacking and finishing a camping mech Terran off before his "super mass" push comes out of 3 base camp. I disagree. You're not going to do damage against a COMPETENT camping Terran. Nothing can break a wall of supply depots guarded by vessels, goliaths, and tanks. You have to out-camp him. Defilers. Lurkers. Ling bombs. Queens.

Watch 2869 rageFuryZ meo2T.rep for an example.

In that game, I think I made a mistake midgame. I should have realized that he's hard-core camping, so instead of trying to reduce his army midgame, perhaps I should have just camped at 7 and denied him an expo. Or teched to defiler-lurker earlier and camped at 7.

#2

Hidden expos are underrated. Take your opponent's back nat on Destination. 2868 kas4813T rageFuryZ.rep

#3

I somehow come back from losing my nat. 2866 Lesley89T rageFuryZ.rep

Just played one of the most epic games ever. Clocked in just under an hour.

Yea, if I defended 5 o'clock it would have been easy for me. Yea, he forgot to land his twin dropships. Yea, I was careless about letting his vessel count grow, etc. etc. If either of us played well it could have been over much sooner, but it wouldn't have been as epic.

Game goes down to the wire, distance mining, vessel sniping, burrowlings, oh my -- I woulda been seriously pissed had I lost.

Me vs. MorroW

/brag

Ever wonder Who is Who in ICCup Season x?

Ladies and gentlemen, please allow me to present this to Starcraft enthusiasts all over the world. I present it with the wish that the original Starcraft lives far beyond the inception of Starcraft 2 and that we will have many more chances in the future to utilize this tool.

I've mentioned in my previous blog posts that I've been working on an automated replay identifier. I've used Support Vector Machines and sophisticated machine learning tools to tackle this problem. However, I've re-evaluated the problem and I've come up with something simpler. I gave up trying to program a machine that is able to identify a player with 100% accuracy. It now presents a list of results that the human must interpret.

Throughout the process, romad has always been a source of inspiration, though he is probably not aware of it. He is the proof that a replay identifier is possible. Perhaps his ability to identify who's who in a replay is nothing short of genius, something akin to humans recognizing patterns in a game of Go. However, with the hopes that his ability can be turned into an algorithm, I kept myself on track.

I present to you INTREPID.

intrepid

characterized by resolute fearlessness, fortitude, and endurance

It is the INTeractive REPlay IDentifier.

Now, I do not claim that its accuracy comes anywhere near the identifying prowess of romad. In fact, it currently does pretty damn terribly.

Where it's at
=====================================================
=====================================================
>>>> INTREPID <<<<
=====================================================
=====================================================

It's in its infant stages and it works somewhat decently. Because of its inadequacies, it is currently only very good at identifying certain players.

It's quite peculiar. It's good at identifying players with very unique signatures, such as IdrA. He has an incredibly unique signature.

I am completely confident that it can detect IdrA replays. You can download the recent IdrA vs. ret replay and upload it to see the result: http://files.getdropbox.com/u/4422/replays/sleepwalker.rep.

I'll go through the process in case you're confused at the results. As mentioned earlier, the results require human analysis.

1. Upload the replay. You'll see the following.

I'm making an automated replay player identifier. With this, it'd be much easier to tell whether you played against a progamer or a well-known foreigner on ICCup. With this, romad doesn't have to be working all the time.

For the time being, I am not going to pursue the Support Vector Machines approach to classifying replays, because we already have an instance that can learn a replay signature just by looking at one sample (romad). I took it as proof that it is possible to formulate a much simpler replay classifier, and that it is the feature selection that is prominently the hardest task.

Support Vector Machines might be more accurate in the long run, after I amass hundreds of samples per player, but realistically, the main problem with identifying replays is their scarcity.

I'm using the Nearest-Neighbors algorithm to classify replays, with a slight modification. I am currently designing it to be half-human half-machine.

There are currently five categories, the fifth being the combination of previous categories.
1) Building hotkeys
2) Hotkey actions
3) Hotkey spam
4) Hotkey usage
5) Total

The program produces a "closeness" rating for a given replay in each category, and looks for replays in the database that are closest to the given replay.

For the database, the program uses replays from TSL (there are a ton of replays from TSL!).

Example
I put in the recent ret vs. IdrA replay. Here are the results:

1) Building hotkeys
Top fifty consists entirely of IdrA.

2) Hotkey actions
The top 4 consists of IdrA.
Top fifty consists mostly of IdrA.
Top fifty has many ret's.
Top fifty also has BRAT_OK's and some Haypro's.

3) Hotkey spam
EDIT: Almost entirely of IdrA. I had a bug in my previous program.

4) Hotkey usage
Lots and lots of Horror. Some IdrA's. The thing is, the top fifty have NO differences from the given replay. That is, hotkey usage is a pretty poor discriminant. I'm sure many IdrA reps have the same hotkey usages, and that if I listed the top 100, IdrA would appear.

5) Total
Top fifty all consist of IdrA's.

It's clear from this example that often, the player will not win in every category. However, put together, it's likely that the player will be at the top.

I will be adding / removing / refining categories in the future.

Also, eventually, there will be an online database and an online real-time classifier, for which I need the Linux version of the repasm.dll that serves as the PHP extension in Windows (I'm using Taiche's RepASM).

Anyway, please send me replays of foreigners if you'd like to test my current program!

So it was around midnight and I had jumped from 7700 to 8200 points. The 32nd place guy had around 8400. I wasn't aware of this at the moment. I dropped down to 7900 with two hours left in the tournament. I lost again and went down to 7800. I realized I needed at least five wins in 2 hours.

Could I have done it had I really tried? Maybe. I wasn't finding any games and I reasoned that I couldn't win enough games. So I gave up not wanting to disappoint myself after going in too deep.

Anyway, here's my Valor tournament rep-pack (majority of games are ZvT): http://dl.getdropbox.com/u/4422/replays/stryker%20val%20reppack.zip

Recommended games (all ZvT) (and approximate replay numbers)
v LastShadow (#0827)
v Heme (#0843)
v Daekyu (#0858)
v worldfamous (A- Terran) (#0909)
v Advokate (#0912)
v Napoleon (#0913)
v kisslyn (A- Terran) (#0945)
v Tarson (#0988)
v IFU.Lucky (#1035)
v Odysseus (#1090)
v Tidy (#1098, #1104)
v Oystein (#1140)

Owned
That is, I'm owned.

When I wrote the first log, I was at 46-9, ranked B- with 6307 points.

As I am writing this log, I am at 81-45, ranked B with 7296 points.

How I Got Here
As with before, I played only ZvTs, my best matchup.

But certain things happened. MOTW kept changing, so I had to play Blue Storm for a week and then Colosseum. Supposedly, Blue Storm favors Zerg, but I couldn't tell. I got owned on that map. My rec is 8-9 on that map. Then Colosseum. Oh boy, did mech kick the shit out of me on that map. Went 14-15 on Colosseum.

Now the MOTW includes Destination again. Destination is my favorite map. Or so I thought. I went 46-9 on Destination till C+. Against B, B+ level players, I'm 4-10 TT.

Notable Foreigners
The coolest thing I'm getting out of this tournament is playing notable foreigners.

Tidy 2-2
Tarson 2-4
Napoleon 1-1
Advokate 0-1
Undergrounder 0-3 (don't know his aka, he was good and he's from Russia)
IFU.Lucky 2-1 (notable? maybe not)
AccordV6 0-1
SrControl 1-1

Gosu Koreans
They're freaking good, especially at B level. I ran into two or three that I deemed untouchable. I mean that I thought I had absolutely no chance even had things gone my way. The untouchables were superior in macro, micro, you name it. It was a feeling of hopelessness that told me to quit SC.

Here are some replays of me vs. Koreans:
odysseus (I'm 2-5 against this 16-year old. He 8-rax rushed me to death 3x though. Me and my noob 8-rax defense.)
EPIC WIN http://dl.getdropbox.com/u/4422/replays/notables/1096%20odysseusT%20VALStRyKeRZ.rep
EPIC FAIL http://dl.getdropbox.com/u/4422/replays/notables/1094%20odysseusT%20VALStRyKeRZ.rep (damn long game TT)

onesoldier (This kid is just sick. His multitask >>>> mine.)
http://dl.getdropbox.com/u/4422/replays/notables/1088%20VALStRyKeRZ%20onesoldieT.rep

NameAcademy8 (There's a bunch of these IDs. NameAcademyX. I think they're all part of some guild or clan or something. They're all pretty good. Amateurs in training?)
http://dl.getdropbox.com/u/4422/replays/notables/1058%20VALStRyKeRZ%20nameacadeT.rep
He does a mnm + fast valk for muta defense. I think this build has gotten more popular nowadays. I believe the correct counter is to take out the valks and score points with mutas. They'll have at most one group of marine+med and two valks, so if you know they're doing valks just make some scourge and harass with mutas like your life depended on it. If you leave them alone you're behind.

Oh, I did manage to beat an A- Korean Terran. I beat his cheese

Brag replay: http://dl.getdropbox.com/u/4422/replays/notables/0909%20worldfamouT%20VALStRyKeZ.rep
His rank: http://www.iccup.com/gamingprofile/worldfamous.html

Keep Going?
I don't know if it's worth my time to try anymore. The cutoff for top 32 in VAL is going to be like mid B+. If my win rate stays the same I'll barely get B+ by the end of the week. TT People are trying so hard :O Should I give up?

OMG
It's not been a week and I've played over 50 games. I have never mass-gamed this much.

O_O

How I got there
My rec now is 46-9, ranked B- with 6307 points.

I got a loss early on against a Zerg player on Destination. I picked Terran (offrace) because I thought I could 8-rax and then 1fact for a quick win. Turns out I was against a B+ Zerg just starting at D level. TT He went 9-pool and owned me hard. Then I won like 10 games straight. I had some near-misses due to carelessness but muta micro owned everyone.

Then I lost against two Koreans with super mech builds. :O Then I got brilliantly cheesed. Then won 10 games straight. Then started losing a couple.

I'm currently 3rd, right behind lastshadow. I played lastshadow once and beat him actually. He faked a mech build and used a fast tank rush on me but I muta-micro'ed for the win. I was thinking, 'I thought lastshadow hacked and sucked without hacks', but he was actually kind of decent. His strategy was a fail against mine, but I imagine with a good strategy he might actually have a chance against me. He gg'ed as well, showing some manner. I was impressed. Lastshadow was not a total ass that I thought he'd be. I'm pretty confident I can beat him if I met him again.

I got to where I am by playing mainly ZvT's, my strongest matchup. I've played 40 or so ZvT's on Destination. It's possibly the most fun map ever.

Why ZvT on Destination is fun
There are so many strategic options for the Terran. Even the 8-rax rush can be executed in so many ways. In between nat and ramp-expo, behind the main, in between the main and the nat, in the middle of the map, in his ramp-expo... The possibilities are endless. Then there is the follow-up. 1fac1star, 1fac expo, 2fac, 1fac2star, 3rax all-in, mine the mineral behind base and sneak vultures in, push marines and vultures and sneak a vulture in, sneak a 5hp vulture in and kill 15 drones.

It's so fun having to think strategically and guessing what the Terran is thinking. I must have been cheesed in every possible form. Every time something new came up I usually lost. I've seen a BBS build that was pretty clever. Against good players, the mind games are not very typical. They'll make sure you get tricked and even if you see it, they thought it through, so they have very good follow-ups. Against bad players, it's easy to spot a cheese and their follow-ups suck.

I'm not being sarcastic at all actually. It's fun having to counter these various strategies and actually having to think during a game.

My ZvT is B+ level?
I beat some good players who topped B, B+ in previous seasons. When I won, I had good control of the game and had a feeling I'd win, so I don't think it's luck. Plus, it's happened a number of times. Once I get to Hive against non-mech, I usually win the game. That's usually because of my muta harass, which has improved recently. I'll kill a bunch of scvs, a bunch of marines, a bunch of turrets, and pretty soon I'm ahead. Stall with lurkers, expo, stall with defilers, ram with 12 ultras and win.

Mech is fun
Mech builds are fun to counter. I love the moments when a vulture tries to run by my sim city thinking they can squeeze between two diagonally placed buildings but of course they can't so they do a dance before the sunken colony takes them out. Sometimes they'll do a fantasy build and do early vulture drop but I'll block it and go in with 9 mutas and there's no defense at home. Good players will have some golies and turrets and (gasp) a Valkyrie. Valks are kind of tricky to deal with. I can't do damage I want because of them. But usually Terran builds too many, like 3 or 4, so I just mass Hydras. They don't have enough tanks and muta + hydra rape gols + valks.

Problems
One problem is handling late-game mech. Late-game marine+med+vessel is easy to handle because I've played it so many times. I don't have enough experience against late-game mech. I just don't know what to do. I expo, I build mutas, build hydras, but they mass golies, tanks, and vessels and just roll me. I try to counter but that usually fails too. I'm going to have to study that a bit more.

Not enough games at B- level
It's much harder to find games at this level. Good thing is I made a friend who I can play. He's at B- and he's okay. Also, I might have to play zvz and zvp because of the lack of games. O_O. TBH, I think my zvz is C level and my zvp is C+ level.

But as long as I can find Terran opponents I think I can get to B+ before the end of the Valor tournament.

Replays
I'll release all my Valor replays later, but here's a fun one.

This Terran basically pulled all the TvZ cards that you see on Destination in one, giant, long, strategy. Everything's there except the 8-rax. He goes two fact vultures + speed for run by, makes a wraith, upgrades mines, makes dropship for 4-vulture drop, and builds a valkyrie for muta defense. But Starcraft is fair. If you can do all these things, then necessarily, your strategy must admit a weakness: http://dl.getdropbox.com/u/4422/replays/notables/0854%20VALStRyKeRZ%20vvT.rep

Short ZvT on Destination, vs Transcendence
http://dl.getdropbox.com/u/4422/replays/iccup%20complaints/0842%20VALStRyKeRZ%20TranscendT.rep

Post-game messages:
"gj you hacker"
"I'm going to report you"
"no one scouts there"
"my main is Protoss"
"I'd own you ez"
"you don't know the game son"

Funny thing is, had he concentrated on the game, he could have won.
I reported him for leaving before 2 minutes. He will be punished soon.

UPDATE
Hi,
I watched the game and here are my changes:

Transcendence -100 point

Best wishes,
ICCup.NeverWin


Long ZvT on Destination, vs Egypt77
http://dl.getdropbox.com/u/4422/replays/notables/0843%20Egypt77T%20VALStRyKeRZ.rep

Post-game messages:
None

Conclusion: Better players are usually well-mannered (notable exceptions include HanDy and Idra).
Also, don't disc-hack me. I WILL see the end of it. Here's what happened to someone who disc-hacked me: http://www.iccup.com/gamingprofile/clen..html

Recently, I've been looking at Starcraft through a different lens. What I mean is that if you take out the micro component of Starcraft, it becomes a turn-based game.

Zerg player opens with a 12-hatch.
Terran player opens with 9-supply 10-rax.
Terran responds to Zerg player with fast cc.
Zerg player builds 3rd hatch, also starts Extractor.

It's like both players respond to each other turn by turn. The approximation is not completely accurate by any means. Neither player has to wait for the other player before making his turn. Also, a player is not limited to performing one act per turn.

However, with this perspective, I found the game to be more about psychology and strategy than micro and mechanics. It's nothing novel or groundbreaking. It's just a change in perspective.

But it's allowed me to finally break the C+ barrier and enter into the B- world.

I used to focus mainly on my micro. I made custom maps so I could practice muta micro. Three Terran bases on Python with turrets all over. I'd have to destroy all of them with only mutas before they built enough to run me over. I focused on ling micro and getting my timings exactly right. Soon, I had mastered the 3-hatch spire build. I would blindly go 12-hatch 11-pool 14-hatch and tech to lair. I'd build drones until 27 and build a couple sunks or lings. My final goal was to get Hive and then use my economic advantage to seal the game with Ultralisks and Defilers.

It was so formulaic and repetitive that I could play without thinking. 12-hatch 11-pool 14-hatch. Micro mutas. Go to Hive. Get Defilers. Get Ultras. Expand. Defilers. Expand. Ultras.

My average APM soon broke 250 and I was regularly hitting 270 in ZvT. But I was still losing games. I'd do 3-hatch mutas but they'd have wildly varying returns. Through D+, C-, and C, I'd easily win with my formulaic muta->lurker->hive. But by C+, I would lose some games and I couldn't find out why I was losing. I thought, 'I'll just have to work on my micro a bit more.'

And I did. My muta micro improved by a significant factor and I could end many games at C+ before Lurkers. The beauty of Starcraft is how important mechanics are. But I still kept losing. I finally realized that I needed to actually THINK during games. Yes. Starcraft isn't just a formula. It's like chess and go. I had to THINK about different strategic options instead of just blindly following the 3-hatch build.

That's where the turn-based perspective helped. I looked at it as a tennis match. I had to react the best way I could depending on the other person's discrete moves. He opened 8-supply? Well, then his barracks will be early so I'll go 12-hatch 11-pool. He opened 9-supply? Well, I will go 12-hatch 13-pool. I soon developed a sense to know exactly if he 8-supply'ed or 9-supply'ed. I found that when my drone scout found the barracks being built and it was only half done when my 12-hatch was about to begin, then it was a 9-supply. I'd take advantage of this to the fullest, often without the opponent even knowing that this happened.

I started thinking a lot more in these terms, in terms of responses to the other player's strategies instead of obstinately following my own. It was about taking turns and whoever fails to respond properly loses the game.

I remember a long time ago when I opened my typical muta opening and got run over immediately by a sunk bust. I remember thinking, 'I needed better muta micro.' How naive. It was my lack of a proper response that lost me the game, not my micro.

More recently, I scouted a 4 barrack build by the time my spire was done. I immediately canceled my expo hatchery and started a few sunks at my nat. My lurkers would never had made it in time to defend my expo against his marine medic mass. However, I also knew that there was a weakness to his early 4 barrack move. Upgrades, tanks, and vessels would be slightly late. Therefore, my late expo wasn't as big of a disadvantage so I didn't have to resort to an all-in and just play safe from there.

These are the kind of strategic thoughts that I have, thinking about the game as a turn-based game. Because I worked so much on mechanics up until now, the execution of my response comes second nature to me. It's just the responses themselves I have to improve. I'm glad that I started with mechanics and am now working on improving strategy, although I can see someone working it the other way around.

There's definitely something to be said about strategic masterminds like Boxer and Savior. It's something that's not evident because it's not flashy. Two marines killing a lurker; two tanks with a dropship killing 4 dragoons. A real game is rarely won solely by these flashy tricks. These tricks only confirm the strategic advantage already gained by the player doing them.

Now I actually think in my games. Strategic responses to the other player's moves. No two games I play are the same anymore. I slide in an extra drone before my pool if I can get away with it. Even a 12-hatch 14-pool is possible. Little variations due to what I perceive to be a correct response to the other player's move.

Jaedong seems to be especially good at countering someone's style. His game flows like water; against Fantasy's mech build, he kept going 2-hatch. Against a marine+med strategy, he goes 3-hatch mutas. Against Zergs... well, we know how Jaedong does against Zerg. He once said that he doesn't really practice specific openings vs. Zerg, that he just kind of wings it depending on the opponent. I believe that is the biggest difference between him and other Zergs. Other Zergs prepare specific builds while he goes into the game and counters the strategy on the fly. His APM and mechanics make his executions perfect.

EDIT: Some sample build orders you can play around with depending on the opponent's card.

I realized my old, "by the book" 9 mutas hatched at 7:15. After some tweaking, I got 6:55. After my change in perspective and realization that there are many things I can take advantage of, I can hit 6:40. The seconds matter because 20 seconds is the difference between 10marine2medic and 15marine4medic. You want to get mutas before they can attack your sunks. If your mutas are late, you're forced to build sunks.

12-hatch 11-pool with extractor started when third hatchery = 6:40
(very unsafe econ though, if terran pulls some gimmick your eco will be terrible due to extra sunks) --- a safe opening against everything

12-hatch 13-pool 14-hatch with extractor started with third hatch = 6:40
(pretty safe econ, basically a richer version of above) --- a safe opening against 9-supply 1rax, or a terran you KNOW to be un-aggressive and won't bunker rush you

12-hatch 11-pool 14-hatch 16-gas = 6:55
(safe econ, but slower than above. basically a slower version of above. the "textbook" opening I learned a long time ago) --- safe opening against everything

12-hatch 13-hatch 15-pool with extractor started with pool = 6:40
(monstrous econ, but obviously dangerous in the beginning due to pool after 3 hatch) --- dangerous against anything other than 14cc

As you can see there are advantages and disadvantages to each build. Depending on the card that the terran plays, I can utilize what I want. I think it's important to know the strengths and weaknesses of the opening you use. For the ones that have super rich econ, I often build a fourth hatchery in my main before spire finishes (because I know I'll have 1200 minerals by the time I have mutas). For the ones that have weak econ, I know that every sunk delays my timing by a huge factor. So I use the card that I believe is the best for me.

I almost always upgrade Zergling speed after Lair vs. Terran. It forces the Terran to think twice before being aggressive, keeps him in his base sometimes, and I often don't have to build any sunken colonies.

The strategy of getting ling speed is suggested by the two following Starcraft theorems.

1) The best defense is a good offense.
2) Army power decreases with distance.

I build sunken colonies as a last resort, if I lost all my lings or I screwed up my economy too much to afford using larvae for a dozen lings. They don't move, and all they do is force the Terran from attacking your base.

Speedlings weakly dominate sunkens because they can achieve the same effect (marine+med don't attack you) but you can do more (like counterattack or harass). The only downside is that instead of using 3 drones to make sunkens, you need 6 larvae (which could have gone to 6 drones). But for the reason I mentioned, I almost always get speedlings because of their mobility.

The second point illustrates that if I have lings, I'll be able to camp in front of his base with them before he can camp in front of mine. By the time he pushes me all the way back, I'll have mutas and his army will be useless. Keeping a distance reduces the power of his marine+medic army.

Today I realized another benefit. Speedlings allow you to take advantage of greedy builds. Builds that, if you don't see it and deal with it when there's a timing window, will roll you over for sure later in the game. E.g.: Terran goes two engineering bay before academy for dual upgrades; Terran goes 1rax1fac1star for REALLY early Vessels.

Here's the recap of what happened. It's Destination, I'm Zerg, he's Terran, I'm at 5 o'clock.

My initial ovie is sent in a dumb direction.



Later, I am able to scout first his Starport.



At first, I think he is just going early Vessel. But then I see this.



He was gonna go early Valkyries, and there would be a timing window in which he'd only have one Valkyrie and a handful of marine and medic. If I left him alone, he'd skip making turrets. Valkyries are to turrets as lings are to sunkens. I couldn't let this happen.

I had ling speed. So I hatched about 6 more lings for a total of about 18.



As soon as I had 8 mutas or so, I go into his nat.

I knew that given his build, he could only have had a handful or marine+med and a Valk. I was confident in the efficiency of my build so that I could apply the substitution principle to the enemy -- you can't have everything; for everything extra you gain, there's something you lose. He sacrificed marine count for early Valkyries, and I was going to make him pay for it.



I trust myself and go in, killing about a dozen SCVs, doing severe damage. In retrospect, I think I could have done more damage by going into his main and attacking his Barracks.

That's pretty much the defining move of the game. The rest of the game is me trying to throw the game by getting constantly harassed. You wouldn't take away much from it. He keeps making small positive EV (expected value) moves, and I know it. He was coming back. I had been dealt an AK against his A8, but he was outplaying me little by little.







Each picture shows a different hatchery. At least 7 hatcheries were sniped throughout the game.

I had one final chance to set things straight. I went all-in on ultralisks and lings and shoved them in. Instantly, the tables turned and I was able to secure the win.

[BRAG]Check out his record. vooevoli[/BRAG]

Back to my point though -- my observation was that a dozen speedlings provide a very good hedge against a lot of greedy Terran builds.

Replay