UC Berkeley Starcraft Class, Week 2
Thursday, February 5, 2009 at 11:12PM Week 2 of the class was better than week 1 because most of the administrative stuff was out of the way so more time could be spent discussing StarCraft itself.
This week was about units. Professor Feng started by explaining that units are your eyes, ears, and hands in the game. Units give you vision through the fog of war (eyes) and are they are what you use to perform actions such as attacking, repairing, building, moving (hands). What he said we might not realize is that they are also your ears. He mentioned one match where the famous player Boxer put an SCV kind of near some minerals to scout, but it was actually the sound effect for gathering minerals he listening for, rather than the sight the SCV provided.
The next topic was something I refer to as local imbalance, though that term wasn't used in class. Explained in my terms, a game is supposed to have global balance (Zerg vs. Terran for example) but it's supposed to NOT have local balance, or it would be too boring on homogeneous. StarCraft has massive local imbalance amongst units and that is a very good thing.
One example was siege tanks. The longest range Protoss ground unit is the Reaver, with range 8. Zerg have lurkers of range 6. But Terrans have siege tanks with range 12(!) when in siege mode. This is a massive "imbalance" in that it gives you an advantage the other sides don't have. That is not to say there is any design problem (it's a strength of the design, not a problem), but it is to say that you have to recognize what exact situations and unit combinations are completely slanted in your favor so you can play to get into those situations.
Feng then asked us to name any Terran melee unit. Firebat? No, it's splash damage. SCV? No, they can't attack under Dark Swarm. There are no Terran melee units, he said. This was just to show that local imbalances swing the other way if you need to fight on the ground under a Dark Swarm.
Next, he asked us "What unit are you afraid of from Protoss if you are playing Terran?" Someone immediately yelled out Carriers. He then explained how good a cloaked Wraith would be against Carriers. Of course, your opponent knows this so your opponent has Observers to reveal the cloak of your Wraiths. He then examined the Medic's abilities, explaining the function of Flare. Flare blinds units and also turns off their detection abilities. Putting all this together, he said, you could use Medics to Flare all the Observers, then attack the Carriers with your cloaked Wraiths.
Here is a video of Boxer doing just that. (Is this a Boxer appreciation class?) Notice that he actually sends in just one Wraith initially, as a test of whether the Carriers can see it. They can't so he sends in the rest.
Feng then talked about damage types and sizes. As you probably know if you've played StarCraft, units are classified as small, medium, or large and attacks can have classifications of concussive (good against small, bad against large), explosive (the reverse), or normal (full damage vs all sizes). He then pointed out that Zerg's Sunken Colonies deal explosive damage and asked us to think about the implications. Most early units are small, while later game units tend to be bigger. That means Sunken Colonies actually increase their effectiveness as the game goes on. (Yes I know that's a simplification because of things like air units and siege tanks, but you get what he meant.) He said we should note though that units can upgrade their damage (zealots for example), while turrets can't (they only get a defensive upgrade).
Feng then asked us to think about how seeing some of the enemy's units gives you information. Not only what they attack you with but what they DON'T attack you with. A specific tell to look for is how much gas their units cost. You usually have some idea how much gas the opponent has, and if he sends wave after wave of units that don't cost any gas, then you can guess what's next. Feng asked, "If your opponent sends wave after wave of Zealots at you, what do you think is next?" One student answered "more Zealots" and I thought to myself that would make a good Rose player in Street Fighter Alpha 2.
Next, Feng asked us what would you do if you've reached late game, the map's minerals are mined out, your only units are a single drone, some overlords, and one defiler. Your opponent has one Wraith. Furthermore you have only 47 minerals, and 38 gas. One student said "I would type gg," which Feng accepted as usually a good answer in a case like that. But he showed a video of just such a situation.
The Wraith attacked the Drone. In moments, it would die. Before it did, the Zerg player put the Drone inside an Overlord, just to buy a little bit of time. He then used the Defiler's Dark Swarm many, many times in a row as a way to transport the Drone safely to a gas mine. With swarms everywhere and temporary protection from the Wraith, the Drone mined enough gas to build Scourge, then used the Scourge to kill the Wraith. This made the game a draw instead of a loss. Here's the video:
Someone asked how this happened and Feng explained that the Zerg player also had a pack of Ultralisks, but they were worthless because the Terran could fly his buildings around, preventing the Zerg win. The Defiler actually consumed one to keep up the Dark Swarm barrage.
Yosh then took over and showed us several examples where a player attacks in a way that is designed to get an advantage but is specifically not intended to win the game. The first was a reaver drop with a shuttle, 1 reaver, and 1 zealot. The shuttle landed the enemy's base right by the minerals, and adjacent to the enemy's one siege tank. The tank went down almost instantly, and the reaver tried to kill as many SCVs as possible while the Terran player scrambled to move them to safety. I cringed as each SCV died, because losing any of them, not to mention all the time lost where none are mining, is deadly in a game of exponential economy. Several died, at least 5, before the Terran was able to surround the reaver with SCVs and kill it.
Yosh then asked us if the Protoss player intended to win the game with this move. It's certainly possible to, and it's been done before. But we know that he really didn't because at the same time, he was building an expansion. Yosh said that if the Protoss player was serious about winning right then and there, he would have invested in more attacking units, send those too, and not put the resources into expanding. The player's actual choices were very good though. The attack did serious damage to the Terran while leaving the Protoss player with good economy.
The next example was a Protoss player who sent 2 corsairs around the map. What was his objective? Obviously not to win the game on that move, but to scout. It wasn't only scouting though, Yosh explained. The corsairs were able to fight a couple overlords, so the Protoss player gained more than just scouting information. Also, the Protoss player knows that one of Zerg's strong moves in this match is to build a bunch of Mutalisks. The Corsairs were really looking for a Spire to see if the Mutalisks were coming or not. If they *were* coming, then having corsairs is a already a headstart on countering the Mutas, so he thinks that's what the Protoss player was gambling for. No Spire was seen, but it seems that the Protoss player's actions were sound and his gains were good.
Next he showed us an example of the Mutalisk stacking bug. By making a group of Mutalisks with one (far away) Overlord in the same group, a bug in the StarCraft game engine makes the Mutalisks stack even more exactly on top of each other than they normally would. This also lets them attack simultaneously, as if it's one big attack. Yosh showed us a match where a Zerg player sends his stacked Mutas to the Terran base. Does he expect to win off this attack? No he doesn't. What he finds is a pretty well defended Terran base with turrets.
The Zerg player circles the Mutas around the base, going for containment. The Terran will not be able to leave the base while the Zerg player expands to a second (or even third) base. But as the Mutas fly around the Terran base, they find an opening and are able to take out a few Terran units. The match goes on to show some cat-and-mouse play with the Mutas, some are lost, but the point is made. They harassed, they contained, they destroyed some units, and they allowed the Zerg player to expand without trouble.
Yosh's final example was a match where the Terran player sent 3 Vultures and 3 Siege Tanks to the enemy Protoss expansion base. Before doing this, he scouted and gathered enough information to believe that attacking early would probably be favorable to him. He did not know exactly what was in the Terran main base though (which was up a ramp). Anyway, he sent the attackers, clearly having Vultures in assigned to one hotkey and tanks to another hotkey. As he arrived, the Protoss player pulled all his expansion probes away, up the ramp to safety while mobilizing 4 Dragoons to fight the Terran units. The Terran player quickly laid spider mines everywhere to control space around the expansion, then attacked the Dragoons. The Dragoons were forced to retreat up the ramp to the main base. The Terran was then able to destroy the expansion base.
Yosh said the whole thing was executed very well by the Terran, but he pointed out two improvements the Terran might think about. First, if he had driven his Vultures ahead quickly at the start of the skirmish, he could have laid the spider mines in a way that would have basically forced the Dragoons to get hit as they retreated. The vultures would then have been able to actually kill some of the Dragoons. Second, Yosh believed that it would have probably been a better idea to bring 3 SCVs to the fight. At the very least, these SCVs could repair the tanks during battle. They can also act as a wall to buy for the tanks as the slow dragoon shots hit them (or float around them due to strange bugs with Dragoon shots). Also, the SCVs could then stake out the territory by building turrets or whatever else.
Anyway, the Terran player didn't bring SCVs and what he did after the battle, I found very interesting. He left. That's right, he just turned around and went back to his base. I have spoken much about "pressing the advantage," but in this particular situation, it does seem that leaving was a smart choice. Yosh referred to this as "accepting the advantage." The Terran could have pushed up the ramp and tried to win right there, but it's really uncertain how that would pan out. He could have lost all those units that way, or he could have gotten lucky. He didn't have the main base scouted and even if he did, 4 dragoons and a ramp can go very wrong for you. Instead, he let the game play out, knowing that they would end up in a 3 base to 2 base situation, with Terran advantage.
Finally, Feng took over again and showed us a video about how 3 Marines can cleverly avoid all damage from a Lurker by making the Lurker attack, then moving to avoid the linear attack, then briefly firing and repeating. This is very, very difficult (impractical in a real match to do it to the degree shown in the video) but it shows a concept you can use a little bit here and there in a real match.
He said that studying small details of the game like this is very important and it's how people in the class can contribute to the StarCraft community. As parting words, he gave an offer. Sometimes when a Dragoon shoots at an SCV, the shot misses completely. Maybe it's something to do with the SCV turning at just the right moment, he doesn't know exactly. He said if someone can show a way to make a Dragoon miss at least 50% in this situation, that he will give that student a passing grade for the entire class.
That's all for this week, and it was a lot. Did you actually read it all?
Reader Comments (67)
Referring to Boxer supposedly listening for mineral mining sounds with an SCV--I don't quite recall the range you need to be to hear sound effects (it's around visual range yes), but there is a very related trick in common usage for scouting in 3 or 4 player maps where you're not sure which spawn location the enemy got at the beginning. My guess is that Boxer actually could have been using this rather than the sound.
Each race's workers have the same sight range. When you have sight of a mineral patch, you can click on it to reveal how many minerals are left. Since enemy workers mine from one side of the minerals (for this example let it be the right side) and the minerals are a couple blocks wide, if you move your scout worker from left to right to just see the left side of the minerals, you can get sight range of the minerals while still being out of sight range of enemy workers. Thus, you can check if the minerals have been mined yet without them knowing you're scouting their location.
Usually this is done with a very early scout to determine enemy starting location. Then you can build early production facilities in the opponent's base or close to it potentially without them knowing you've even scouted early, for a high risk high reward rush build.
Well, Zerg on many maps does the "scout the minerals" trick frequently using first overlord. Same principle as worker scouting but early worker stealth scouting usually implies a rush build whereas overlord scouting does not.
I'd take a Street Fighter class taught by Sirlin.
Rekrul and a lot of the other other top SC players like the APM intensive style of play. He also confuses execution with strategy.
Don't take what he says to heart. You should see the big flames they have on teamliquid against the modern UI updates in SC II.
For me, I love SC for the wide variety of strategies, and the beautiful way things can be pulled off. I admire the pro apm level executions, but I enjoy watching the strategies they use and different tactics they come up with more than admiring their ability to click fast.
I'm all for fast games and everything, but Teamliquid players usually take pride in SC's outdated interface and controls. If they had their way, SCII would be the exact same thing as Broodwar except in 3d and maybe some new units.
Don't pay too much attention to him. Your articles are insightful and interesting to read.
I have been reading a lot of these comments, and i would like to state my views on this subject:
I AM a frequent visitor of teamliquid.net, because they have an awesome staff, AND some really smart and nice members. This might not be apparent at the thread most of the visitors are looking at, but that is because this subject really brings out the very very worsst of the members.
Why is that, you might ask. It is so, because TL.net is split between players wanting a very strong UI, meaning, as it is now in SC2, workers go mine automatically when produced (automine), multi building selections (MBS) and autosurround in attack move (go look up some reports from BlizzCon at TL if you want to know more).
The other faction of players want something very close to the old UI of SC1, because they believe it is the best way to keep mechanics an important part of the game.
Now, here is one thing i want to stress: Almost all of TL.net does want mechanics to STAY a part of Starcraft, because they DO agree, that it is a necessary part of the game.
Okay, just wanted to give you some insight on what TL is, and what it isn't.
Now on to what I think:
I believe, that mechanics are a necessity in both SC and SC2. The timestrain and frantically clicking everywhere, brings two whole new resource into the game(which, IIRC, Sirlin mentioned in Week 1): Time and attention.
If these two resources didn't exist, the player who won the double-blind buildorder choice at the beginning of the match would always win, simply because his choice was better. That would be like SF2, where E. Honda would ALWAYS, not just most of the time, but ALWAYS lose to any character with a projectile, and would ALWAYS win against characters without such a projectile.
I am hereby saying, that SC is an amazingly deep game, but it is even more amazingly deep, when you also has to manage your time and attention, because you've never got enough of any of these.
If you took away mechanics, anyone would have all the time they need after very short time, and the game would quickly become very shallow.
So rekrul is basically:
Underestimating the strategic depth of Starcraft
Deriding Yomi as a roll of the dice
Once you take that away his whole argument falls apart.
>I believe, that mechanics are a necessity in both SC and SC2. The timestrain and frantically clicking everywhere, brings two whole new resource into the game(which, IIRC, Sirlin mentioned in Week 1): Time and attention.
SC with a good interface still has plenty of stuff to spend clicks on, like combat micro.
Slivaz: Suppose I were to assert that control groups and shift-queueing should not be a part of SC2 UI. How would you respond to that?
Bill The Maniac:
1. Yomi in this case is easy. Assuming that payout matrix is know you can just use a compy's RNG to choose your build-order at the start of the match.
2. Mirco tricks are not enough to overcome a 'bad' buildorder. You would get like a 5% extra power in combat but that is not enought to overcome the 20% of edge the other guy.
mjw:
Assuming that payout matrix is know ...
How can you assume this? Has StarCraft been "solved"?
Yomi is not really important in SC/BW or every other RTS were scouting let's you actually *see* what is happening.
You allways want to know what your enemy is doing, Yomi is a guess, you need it if your scouting fails. The more you rely on Yomi the worse it will go for you. Even if your a genious at *feeling* what your enemy is doing, scouting information is a 100% fact. If you get good scouting in you don't need Yomi because there is no reason to *feel/think* about what your enemy is doing, you know it because you see it.
Facts > Yomi.
In Street FIghter or other games this is an entirely diffrent case because you can not really *scout* what your opponent is up to, you will notice it when he actually begins to show you his gameplan, you have no chance to see it before he actually starts it, you also can't see a *strategy* switch until it's done. Yomi is important in such a game because you have no better options.
Yomi is guessing. If you can have *facts* guessing fails.
mjw:
Assuming that payout matrix is know ...
How can you assume this? Has StarCraft been "solved"?
I'm only talking about the opening double-blind pharse of the game. The other guy has no way to stop your plan. Due to Starcraft's age the pro's known the payout matrix reasonable well for the opening. (if they did not someone else would-- that person would win a lot.)
very interesting. would love if they had this class here.
craNk:
"Yomi is not really important in SC/BW or every other RTS were scouting let's you actually *see* what is happening."
I respectfully disagree. There are times when Yomi is not important, but there are times when Yomi is very important.
For example, do you send a worker out at the beginning of the game in order to check for proxies? That seems like Yomi to me. You don't KNOW that your opponent is going to build a barracks near your base and proxy rush you, but you guess that he might do it with enough confidence that you send out a worker early on, sacrificing the minerals from the time that worker could have been mining in order to check for something that might not even be there. On the other hand, not doing it is risking that you will get blindsided by a proxy barracks and lose. It's not classic SF2 Yomi, like guessing why Ryu is going to stop throwing fireballs and go in for a throw, but it's something similar.
Another situation is where both players have just killed their opponent's scout worker and have to choose where their builds are going to go from there. In PvZ there is a time where both players have likely killed the other player's scout worker, and they have to pick the builds that will take them into mid-game in the dark. That's the kind of double-blind situation that Yomi frequently occurs in. The Protoss player can choose to build a Stargate and and corsair to scout out the Zerg, but he might not if he guesses that the zerg is going to do something tricky. Say, for example, if the P believes the Z will mass up a large zergling force with their movement speed upgrade, guessing that the P player is going to do a standard stargate -> corsair build, and use the time where the P's ground army is weak (because the P player has been spending his resources on a stargate and corsair) in order to break the front of P's base and either get inside to kill probes or even end the game right then and there. That's Yomi,
Maybe my point did not come across right.
You only use Yomi in Starcraft when you don't have another choice. As soon as you send out your drone to scout you don't rely on Yomi anymore, at least not to a big extend... If you don't send out that Drone (or Overlord) to scout for Proxis and anyway do a *greedy* built, then your betting on your *feeling* that the enemy won't rush.
That’s in the very beginning of the game and in some MU's a little later for a short time again... But in the long term? Corsairs, Observers, Speed Overlords, Mutas, Comsat deliver endless scouting possibiltys.
Aside from the starting BO and a short timed window a little later there is really not much reason to not know what your enemy is doing.
I just don't want to call Yomi a *MAJOR* factor in an RTS when it's just *Plan B*. *Plan B* can win, but you much rather rely on *Plan A*.
"I cringed as each SCV died, because losing any of them, not to mention all the time lost where none are mining, is deadly in a game of exponential economy."
StarCraft doesn't have exponential economy. I know you studied math sirlin, so you should be able to follow the following:
With only one base, the rate of increase in mineral intake is linear, because the rate of SCV production is constant (given that the player produces workers constantly). That makes the mineral intake quadratic.
But that's of course not the whole story. For one thing the intake does not increase quadratically in the case of one base because the mineral patches are finite, so each worker will become less and less effective.
On the other hand one can build more and more bases. That makes the rate of SCV production linearly increasing, and the rate of mineral intake cubic.
This whole agrument is not completely accurate. For one thing the functions in a starcraft game is not continuous (but linear, quadratic and cubic functions are). Furthermore, one can approximate the mineral intake with a cubic function if the rate of expansion is constant, that is the time between expansions is constant (they never are exactly constant, but often approximately constant). And when a player has about 3-4 bases the worker production usually stops because they start to take up too much supply.
My point is that the mineral intake is approximately cubic throughout the game, but approximately quadratic with only one base. If you know anything about complexity theory you know that there is a REALLY big difference between exponential and cubic functions.
Or, I should say cubic until worker production stops, not throughout the game. After worker production stops the mineral intake is approximately a constant function, until lots of workers die or the minerals run out.
I think many people use the word exponential as a blanket statement for 'greater than linear' growth. It presumably entered the general lexicon as a result of interest on banking and loans.
I have never heard someone refer to something as having quadratic growth outside of a math or comp sci class...
A lot of things have exponential growth, like your money on in the bank account, people on the planet, etc.
"I think many people use the word exponential as a blanket statement for 'greater than linear' growth." Yes, it is apparent that they do (and I'm not talking about only Sirlin here :)). But that doesn't make it right. I think it's just ignorance on their part. A lot of people don't know what the difference between a quadraic and exponential function is. Sirlin should know, so I don't know why he wrote exponential. :P
Just something from this week's class that was used as an example where the presence of an APM difference will change:
In the Zerg vs. Terran match, where the Zerg player uses Mutalisks to contain the Terran opponent, it was stated (I cannot tell whether the statement comes from the class or Sirlin, and I make no effort to flame either party) that the intent was not to win on the attack. I just want to point out that a Zerg player with superior APM can, in fact, win the match on the very basis that he is faster by smashing turrets (if he has the Mutas), depleting enemy Marines/Medics, and replenishing his units in a manner where he is gaining Mutalisks rather than replacing them, whereas a smaller APM difference would result in the very same attack as a stepping stone for the late game.
Also, I find that the difference in "accepting the advantage" versus "pressing the advantage" in the Vulture/Tank/Dragoon scenario may not be a difference at all. From my perspective, pulling the units out, which by all means was a prudent choice, was an acceptance of the situation as a moment where it would not result in an imminent "GG". On the other hand, by pulling the units out, the player can focus on increasing his macro advantage because of what he was able to do to the expansion, which is pressing his advantage all the same. May I suggest a more accurate statement of "identifying the advantage", so as to know what one has gained, and therefore take advantage of said gains? Yes, I am nit-picking a bit. No, I'm not doing this to be a miniature chunk of human excrement, but rather to enhance meaning and deliver insight to those who might benefit from it.
I am cognizant that the specific examples shown in the class may have been utilized illustrated otherwise, but I point these out because the exception to the example or difference in perspective were never pointed out in the main article. I am not attempting to flame Sirlin, TL, or Professor Feng by making this statement, so please direct all gunfire at someone else. Thanks.
craNk, after reading your second post, I realized I agreed with you and just didn't grasp what you were really trying to say. My b, homie.
Someguy:
"Also, I find that the difference in "accepting the advantage" versus "pressing the advantage" in the Vulture/Tank/Dragoon scenario may not be a difference at all. From my perspective, pulling the units out, which by all means was a prudent choice, was an acceptance of the situation as a moment where it would not result in an imminent "GG". On the other hand, by pulling the units out, the player can focus on increasing his macro advantage because of what he was able to do to the expansion, which is pressing his advantage all the same. May I suggest a more accurate statement of "identifying the advantage", so as to know what one has gained, and therefore take advantage of said gains? Yes, I am nit-picking a bit. No, I'm not doing this to be a miniature chunk of human excrement, but rather to enhance meaning and deliver insight to those who might benefit from it."
I think you're too caught up in the semantics.
"Pressing the advantage" is just the belief that you have an advantage in your ability to destroy your enemy's base, and that it is the correct move to try and end the game at this moment in the match.
"Accepting the advantage" is the belief that you have an advantage that you can develop over time, and as such you stand to gain the most by sitting back and developing your superior economy; your strength will climb disproportionately to your opponent's strength by containing him and just denying him the ability to expand.
Different situations require you to press or accept your advantage, mdepending more or less on what kind of advantage you have. For example, you should probably press if you have much stronger army but a much weaker economy, because your opponent will soon leverage his economic advantage into an army advantage. On the other hand, your opponent would try to accept the advantage by just playing defense and tryign to win the game later.
Thank you for writing up the summary of the lectures. I enjoy your site and entries, and these were another great addition. Cheers.