UC Berkeley StarCraft Class, Week 7
Saturday, March 14, 2009 at 12:16AM UC Berkeley's StarCraft team defeated MIT's team by a score of 3-2 this week. I think I was supposed to feel happy about that because there I was attending the Berkeley class. But then I was supposed to feel sad because I went to MIT. In truth, I never bought into the us-vs-them concept of school rivalries in the first place but congratulations to the winners.
This week's class was about scouting. Somehow, this involves a bunch of equations and graphs that I don't have time to reproduce for you (nor do I even know how on the web, really) so you will suffer, just like last week. Or maybe you're actually happy about that because it means I'll explain the equations in regular English, so the non-math people will be able to understand more easily?
Choices
Before we get to scouting, let's lay some groundwork. Consider the difference between choices that are discrete and choices that are continuous. If you want to get really technical, just about every choice in StarCraft is discrete because it eventually comes down to the pixel and the clock cycle, but let's not be quite that picky. The categories wouldn't be so useful if we put everything into one of them.
In a discrete choice, we get only the extremes and nothing inbetween. For example, you choose between building a robotics facility or a templar archives. These are different tech trees so this one decision has a lot of implications and you can only have one or the other (early on when you can't afford both). You can't decide to go halfway on that and build half a robotics facility and half of a templar archives.
Continuous choices have much more leeway though. When to make an expansion? There are so many points in time you might decide to do this that it's basically a continuous choice. If you send a probe out to scout, how far to send him? There are many possible distances you could send him out. Another example that I think is important to the overall point is the ratio of your army. Lets say you are building a large ground army that consists of Zealots and Dragoons. Yeah it's technically discrete (so is everything) but it's more useful to classify that as continuous because there are so many possible choices of exactly how many of one unit and how many of the other.
Graphs You Can't See
Imagine a 3D graph with time going up the Z-axis and X and Y being the number of drones and zerglings you have, respectively. At time = 0 (also known as Z = 0), you have 4 drones and you have 0 zerglings. So we can put a single point on that graph to denote what you definitely have. (Remember, you start the game with 4 drones as Zerg.) What about at a later time, higher up on the Z-axis? You might STILL have zero zerglings because you built all drones. Or you might have mostly zerglings and a few drones. Or you might have some mix of those. In 3D space, we can draw a triangle to represent all the possible things you might have. One corner of that triangle would represent you being a total loser and only having those same 4 drones you started with and nothing else. Another corner represents you building all drones, and the other corner represents all zerglings. If you're an expert player, we don't even have to consider the area inside that triangle because it's all suboptimal playing. It's really only the edge between all drones and all zerglings that matters.
So at the exact point in time I was talking about in the previous paragraph, there's a "triangle of uncertainty." As time goes forward, that triangle moves up our graph and expands bigger and bigger. Professor Feng referred to this as that "cone of uncertainty." At time t1, perhaps it was only possible to have at most 9 drones, but at later time t2, you might have 20 drones. And instead of having at most 6 zerglings (or whatever), maybe you could have 24, or something. The idea is that the further ahead in time we go, the greater the possibility space is for what we might have.
Then there were some vector equations basically expressed that what the opponent REALLY has is the sum of what you know they have (because you saw it) plus the uncertainty factor of what you don't know about. The point of scouting is to reduce the size of that cone of uncertainty as much as possible.
Replay 1
Then we watched several replays of matches that show scouting. The first was Protoss vs. Zerg where the Protoss player sent a probe to the enemy's main base. In an early scouting situation like this, the Protoss player wants to know exactly where on the continuum between all drones and all zerglings the Zerg player is. The Protoss player probably wants to know how much army he has to build to survive, and he wants to build no more of an army than that. If he prefers to build all economy for a while, scouting can tell him how much he needs to cut into that for an early defense.
Anyway, the scouting probe first told the Protoss player that the Zerg did not yet expand to his natural expansion. Then the probe showed that the Zerg player is building zerglings. I expected the scouting to end there because, well, we know everything Zerg has now. But the probe ran around and avoided the zerglings for quite some time. Then there more zerglings and the probe still managed to stay ahead of them all, going around in circles. I think the zerglings even got the speed upgrade and the probe was STILL going around the zerg base, using tricks and glitches like passing through a mineral patch. (Only possible when 3 drones are in just the right place mining minerals, you try to mine the same minerals then click "stop" with your probe, then right click behind the minerals."
Anyway, the extreme amount of attention that the Protoss player used shows how important it was to KEEP scouting. The extreme amount of attention the Zerg player spent trying to defeat this single probe shows how important it is to STOP the enemy from scouting you. It turns out that the Protoss player had more attention resources because he had a fairly good economy and decent army building while the Zerg player was so distracted that he hardly had anything and forgot to build one of his overlords on time, constricting his army.
Replay 2
This was Terran vs. Protoss and the Protoss probe reached the Terran's ramp just as the Terran completed a full wall with a supply depot and a barracks. The probe stayed there. Then a single marine came out of the Barracks to fight the probe. The probe left...then returned later and there were no marines to kill it.
The Protoss player is certainly wondering what is up here. His probe was able to go up the ramp just barely enough to get a bit of vision to part of the Terran base. It was the area where a Terran would usually build a command center to fly over to his natural expansion. The probe revealed that there was NO command center there, nothing really. He did see a single tank though (reasonable, as it's necessary to hold the wall.)
But really, what's going on here? The Terran has a quite a cone of uncertainty going on in there. The Protoss player reasoned out what must be going on, mostly from what he didn't see. He saw no attack forces on the way to his base, or anything threatening like that. No expansion, and no command center in the traditional place to prepare to expand. A barracks, but no mass of marines. What's up here? Then he saw Vultures lay a few mines near the ramp with the Terran wall. This was 6 minutes into the game. I'm not sure exactly what that meant, but apparently the point is that those vulture mines are really valuable and you wouldn't waste them there normally. It seems the Terran's only remaining option was to attempt a dropship attack.
The Terran DID attempt a sneaky dropship attack and the Protoss player had already moved Dragoons into position, just as the drop arrived at the Protoss main base. The Protoss player moved all his probes away immediately, lost maybe 1 or 2 at most, and stopped the drop. He was extremely on his toes there and came out of this 2 bases to 1.
Replay 3
Zerg vs Terran. The Terran player scouts early to see that the Zerg player did a fast expansion. This let the Terran player know the Zerg put all his resources into that for the moment, so the Terran player is safe to expand himself. He does. A bit later, the Terran continues to scout with an SCV and happens to encounter a mass of zerglings in the open, on their way to attack. By the time the Zerglings arrive (it was a long distance), the Terran player moved his units around to block the ramp to his main and position a few marines to attack the Zerglings in a good way. Had he not scouted, he wouldn't have been ready. He was able to stop the rush easily though, when knew exactly when it was, where it was headed, and what it consisted of.
Replay 4
In this one, the Terran scouted the Zerg and saw the Zerg's Lair. This signaled to him something about where the Zerg's tech would be. Unfortunately, he did not see what the Zerg's larva were actually building, even though he could have with the scout. The Zerg player actually massed zerglings and attacked the Terran who was really unprepared for this.
Replay 5
Zerg (player is idra) starts by scouting in TWO directions, using two drones. So he's paying quite a bit of economy here to gain this scouting information. Apparently he was afraid of what the Terran might do on this map. The Terran scouts with his SCV and at only 4 minutes into the game, he finds a spire building(!). This means the Zerg player actually hid his lair because the SCV hadn't found it. It also means that there can't be a 3rd hatchery because it's just impossible to have 3 hatcheries and a spire at that point. So what's coming next? Mutas.
The Terran immediately build a LOT of turrets at his bases. 5 turrets at one and 4 at another. They finished building exactly as the pack of mutalisks arrived, and the Terran player easily fended them off. The Terran used comsat to look at the Zerg base and happened to see a queen's nest and a hive building. He also saw one lone mutalisk that had full health and looked like it was probably there because it was at a rally point from just being built.
So the Zerg built a lot of mutas, seemed fairly reckless with them, didn't show any signs of changing his tech. When trying to make a hive, he spent a valuable 100 gas on another useless muta? It can only mean that guardians are coming.
Replay 6
This is on the map called Destination. The main bases have two ramps, but one ramp is blocked by a patch of minerals. This means there's basically only 1 ramp in the early game (possible to defend) but later on there can be two. Anyway, the Terran player sends and SCV up the main ramp of the Terran base (the one that isn't blocked by minerals). The Protoss player very cleverly moves his units to block and kills the SCV without letting him into the base. Even though the SCV saw absolutely nothing in the Protoss base (what is he building?), it did see one thing. It saw the minerals on the ramp. It got vision to the minerals on the ramp. Why does that matter?
The Terran player then sent a second SCV, but this one he use the right-click command to mine those minerals on the enemy ramp. While a worker unit is in mining mode, it can pass through all other units. So the SCV goes up the main ramp (on its way to the mineral patch) and passes right through the Protoss units that blocked the previous SCV! The SCV then went right to the middle of the Protoss base and saw a Citadel (speed zealots coming, perhaps?). The point here is that he knows he won't die to Dark Templars immediately, so it was definitely worth it to know that.
Replay 7
The map is Tau Cross, Zerg vs. Terran. On this particular map, the Zerg player is more worried about a proxy barracks strategy than even finding the Terran base. I think he thought that some barracks placed in the middle of the map was the main thing he might lose to. He sent his first overlord directly to the middle of the map to check for this, rather than the more usual action of sending the overlord around the perimeter.
The Terran WAS building there and the Zerg player even used ninja tactics to move his overlord away very quickly so the Terran would not know his trick had been revealed. The match commentators were going so crazy over this that they said it was "gg" right there (in Zerg's favor) because the barracks were discovered. The class laughed quite a bit at that, imaging their opponents saying "gg" because a single overlord discovered a building, lol.
Closing Thoughts
Professor Feng's last words this week about when NOT to scout. Scouting is most useful when the knowledge you might gain would change a decision you'll make. And even more than that, when the decision at hand is discrete (should I go robotics or templar archives?). If you're already "all in" though, maybe you've gone full force with Zealots rather than full economy, you just need to attack. As long as you know where the opponent's base is, it won't help that much to know which exact part of the triangle of uncertainty he's on. What if you find out he went half economy and half army? So what, you still attack and win.
Of course there is still reason to scout no matter what, and many tricks and harassments you can pull off. But Feng's point is to recognize how much (or how little) scouting will help you at any given moment so you know whether to devote extra resources to it (more than one scouting unit, or more of your personal attention resources) or whether your resources are better spent building a strong army, or whatever else.
Reader Comments (5)
In response to replay 7:
There was a pro game where the Protoss made a proxy gateway in a wrong position, and got his zealot stuck. He typed gg before the other player even knew about the proxy.
If only I could find the vod...
Yet another nice recap. I've really enjoyed these weekly write-ups Sirlin, keep up the good work!
Mmm, hidden information...
I have to wonder though, if and how often players make a point of LETTING their opponents see something strange in the name of confusing him. I would assume that in most of these cases, especially 2 and 5, that the player being scouted could easily fake what he's doing to screw up his opponent's economy. Make it look like there's only one option for what he's doing, get his opponent in a frenzy to prepare for it, and go out and do something else while he's gone crazy with defense against an attack type that never comes.
Granted, doing this would be a rather large risk. You're wasting valuable minerals/gas when they matter the most in the hopes that your opponent will see just enough to get him pointed in the wrong direction. But I would think that this has to happen at least occasionally, and that an example of it should be with the rest of the scouting stuff as a pitfall to watch out for.
[quote]Granted, doing this would be a rather large risk. You're wasting valuable minerals/gas when they matter the most in the hopes that your opponent will see just enough to get him pointed in the wrong direction. But I would think that this has to happen at least occasionally, and that an example of it should be with the rest of the scouting stuff as a pitfall to watch out for.[/quote]
I don't think it's ever a good idea to LET your opponent see what you're building. Think about the cone of uncertainty, in the graph example it's 3 axises but in a real game it'd actually be an n-dimensional graph. At some point in time you could have 0 or more drones, 0 or more zerglings, 0 or more hydralisks, and 0 or more mutas. That's a lot of uncertainty! By letting your opponent see your base, or part of it, you are cutting down on that uncertainty significantly.
A good example of why this would be bad is if you were to fake out a player by tricking them into making a ground firing only unit (like zealots) then rush flying units. For the trick to actually work the units he thinks you are building would have to be strong enough to change his desired course of action. Otherwise the player won't bother to counter your army to such an extent that he's opening himself up to other weaknesses. It'd be an unnecessary risk to go all-in on a defensive counter strategy when a more moderate approach will work.
Since the strategy you are faking is so strong then by not letting your opponent see your fake you can actually execute that strategy and have the advantage. There's no need to fake it out and take a riskier strategy if you can instead prevent your opponent from scouting.
Now, if your opponent does see your strategy early on anyways then you're in a good position to use that to fake him out. Applied to the example above you are going your initial units, the ones that need to be countered by mass ground attack only units, but the opponent spots this early on despite your efforts to stop him. That puts you in a position where you can use it as a fake and go the air units to stay a step ahead of him. Of course then you're getting into yomi so you have to be careful. He knows that you know that he knows you can be building a force that requires a specific course of action to counter so you might instead go the counter to that counter.
It's also not a waste of resources to pull off the 'fake' maneuver if the game continues on. If you pull a fake and attack with the air units and gain some ground. The other player fields his counter to your air units so you're going to need something else at this point, continuing to pump a unit when it's being countered would be silly. The resources you spent on the fake can now be used to counter his counter or throw him off balance again.
Logo said:
It's also not a waste of resources to pull off the 'fake' maneuver if the game continues on. If you pull a fake and attack with the air units and gain some ground. The other player fields his counter to your air units so you're going to need something else at this point, continuing to pump a unit when it's being countered would be silly. The resources you spent on the fake can now be used to counter his counter or throw him off balance again.
-----
If I understand you correctly, is it like this game?
http://www.gomtv.net/classics2/vod/539
There, Flash fakes his "flash build" (Where he gets super upgrades), and goes for 4 or so factories and pushes out.