UC Berkeley StarCraft Class, Week 11

This week's StarCraft class was about deception.

Professor Feng used these terms. leading up to deception:

Level 0 - Use micromanagement, macromanagement, look around the map for various map features, you aren't taking the opponent's actions into account.

Level 1 - Scout the enemy, build correct counter units based on what he has, position your army well given the position and direction of movement of the enemy's army

Level 2 - Deception and reading the mind of the opponent.

Two main types of deception in StarCraft are:

Army deception. Either a) get their defensive army to leave when it shouldn't or b) get their offensive army to attack at a stupid time or place.

Build order deception. Force the opponent to build the wrong units.

We already looked at one example of build order deception in a previous class. The Zerg player used an overlord to scout an expansion base, but was chased off so the overlord could only see the edge of the minerals at the expansion, rather than the whole base. The Terran player then sent SCVs from his main base to mine the expansion minerals to fake that he had an expansion. This caused the Zerg player to expand, thinking he was safe for a while when in reality the Terran had no expansion at all and a sizeable army to rush with.

Here are several more examples of deception. In this video, the Zerg player has 4 lurkers. The natural place to put them is at the top of the ramp, but he splits his force into 2 lurkers at the top and 2 before the ramp. He then puts the lurkers at the bottom of the ramp on "hold" but leaves the ones at the top in their usual attacking state so the opponent will know about them. This coaxes the opponent into gathering a large force at the bottom of the ramp (right on top of the two lurkers down there). Notice the amazing willpower of the Zerg player in the he gets an entire pack of marines right on top of his lurkers...but he still doesn't make them attack. Instead he waits until there are TWO entire packs and uses his mutalisks to bait them into standing in just the right (or wrong!) place.

Here's another similar example, also showing some willpower on the Zerg's part. The Zerg player allows the Terran to destroy his sunken colonies even though the Zerg could take his lurkers off "hold" and have them attack. When the last sunken colony is destroyed, he finally allows his lurkers to attack and they destroy the Terran's entire force (which was like 50 supply or something big). Note that the Terran did not respond once the lurkers appeared. This means that he was not even watching the battle. The Zerg player was hoping for this, and it's standard practice to NOT watch the battle when the result appears so clear cut. The Terran was surely looking at a different part of the map or building more troops or something, so he was caught unaware.

Here's another example of army deception. The first part is just the Terran (famous player Boxer) using vultures to destroy nearly all the worker probes at a Protoss base. The notable thing here is the nearly orgasmic screams of Korean fangirls. Then Boxer sends an (empty!) dropship down the side of the map. He passes a Protoss observer, which is exactly what he wants. The professor said that Boxer couldn't see the observer, but that his extraordinary Boxer-powers allowed him to know that it was there. When the empty dropship reaches land, the opponent counters by having several dragoons ready. Unfortunately for the Protoss, this was a trick that caused those dragoons to get out of position. During this time, Boxer attacked the nearby base with (unsieged) tanks and killed most of those worker probes too.

This video is a bit hard to follow and you probably want to watch it with the sound turned off. Early in this game, the Protoss scouts with a probe and sees that the Terran does not yet have any gas and that the Terran is building a second barracks. This signals what the Terran build is, and the Protoss countered by building a cannon and putting two zealots at his choke point. The Terran actually built the second barracks as a fake though. Instead of going for a quick rush build, he got an expansion (gas from two bases) to fund his tank+vulture army. The only thing the Terran lost here was 150 minerals for the fake barracks while the Protoss lost at least 500 minerals from the forge, cannon, and two zealots--none of which he would have built if he knew the Terran's actual build here.

In this next example, the Terran (Iloveoov) uses an unusual strategy. I think it would fail pretty hard in the hands of most players, but Iloveoov is capable of amazing things. The Terran starts by walling off his own base so that he can build in peace for a bit and so the opponent will not know what he's building. He builds a starport which makes one wraith at first. He harasses overlords well with this single wraith. What's strange is that he also builds an academy and has medics and marines going. The Zerg player sees this and doesn't know what Terran's build really is. Should Zerg build hydras? Or maybe not if marine+medic is the real build here. Or should Zerg build mutalisks and scourge? That could counter air.

The confused Zerg player builds lots of mutalisks, only to find some valkyries(!) from the Terran. Basically, the Terran has all sorts of tech options but hardly any quantity of any given tech. The match commentator points out that this is "playing Terran like Zerg," meaning that the Terran is imitating zerg's ability to switch techs often. It turns out that Terran's "real" build here, the one he finally commits the most resources too is marine+medic with a few tanks. Zerg was so off balance that he floundered most of the match.

This game was the first of a 3-game series. Terran won this game, but he used the same (ridiculous!) strategy the next two games and the Zerg player was ready those times. He just went full hydra because it counters most of the things involved. It would lose to medic+marine, but Terran can't have an early m+m going if he's also building all that other ridiculous stuff.

Next up, one of the craziest plays there is, and of course it's by Boxer. Remember, this is a Boxer-appreciation class. ;) Boxer (as Terran) is in the finals of a tournament here, under huge pressure and with money on the line. He decides to cancel building his supply depot, and instead...fly his command center away. On this particular map, there is an expansion behind the main base and this strategy would only work on this map or a very similar one. Boxer leaves no trace of his original base at his starting location.

His Zerg opponent is scouting to find which of the three possible starting locations Boxer has...but finds him in none of them! Zerg is very confused here and the crowd is very excited. Boxer manages to stop a Zerg expansion, too. Although this looks like Boxer is doing really well, think about it. The Zerg had to cancel his expansion so he's down to just his original base. Boxer only ever had one base though, and really not even a full base because the move put him behind on building supply depots and more importantly, behind on mining. So actually Boxer is behind at this point and soon loses the game. But it's one hell of deception play!

The Famous Boxer

The class administrators briefly mused about Boxer. He was dominant in the past, but these days he relies on ridiculous tricks because his macromanagement isn't what it used to be. They all agreed that the #1 Terran in the world is not Boxer, and yet we all know Boxer's matches way better than the #1 Terran's. Boxer's matches are like shots heard 'round the world because they are so crazy. Then they mused about how perhaps Boxer is the #1 Terran, just not at winning--he's #1 at being known. If you're thinking about my Playing to Wn book right now, remember that Playing to Win was directed at people who want to win but have obstacles in their way. It says nothing about whether having a different goal in the first place is ok or not ok. It's kind of interesting to wonder if Boxer's goal is actually to be the most famous at the expense of winning a bit less.

Call and Response

Professor Feng then talked about the concept of call and response (aka signal and response). We see this same concept in nature in many ways where sending the signal has a cost, but sending it worth it. Male peacocks grow huge feathered tails, so large that they actually drain lots of the male's body resources. This attracts females. Part of the attraction is that if the male is able to spend so many resources on his tail feathers, he must be, well, a baller. (Feng mentioned peacock but the actual explanation on this one is from me.)

Babys crying also exemplify call and response. The baby makes itself weaker with the crying signal because the noise makes it more vulnerable to attack by predators. But parents know this and rush to feed the baby or comfort it so it will stop crying, so it's worth it for the baby to use the crying signal.

The most applicable of these examples to StarCraft is the example of gazelle "stotting." When a lion or cheetah spots a gazelle, sometime the gazelle will jump very high, lifting all four feet off the ground. During a chase this actually slows the gazelle down and allows the predator to close the distance a bit more than if the gazelle did not do this stotting behavior. Even though sending this signal makes the gazelle weaker, it's to the gazelle's advantage. When the predator sees this, the predator knows that the gazelle is strong and in good health. Predators get better bang for their buck by hunting prey that are old, disabled, or slow. So the signal involved in stotting makes predators more likely to give up and find an easier target.

We then looked at an example StarCraft game between two Protoss players. Player 1 has several zealots near his main base. Player 2 has several zealots nearby, close enough to threaten. Player 2 sends a signal with the purpose of seeing if his opponent gives the appropriate response. His signal was to send just two of his zealots past the enemy's pack of zealots towards the workers in his main base. Player 2's response was not terrible, but wasn't good enough. Player 2 sent two of his own zealots to chase down the attackers in the mineral line. This was a better move on Player 2's part than sending ALL of his zealots, because then Player 2's expansion would have no defense at all. But sending only two zealots wasn't quite enough and it took too long to kill the enemy two zealots. He probably should have sent 3 zealots (or 4?).

Player 1 saw this suboptimal response and learned that harassing with zealots would be a good tactic against this opponent. If Player 1 had been stopped easily, he would know that further zealot harassment here would not be worth it (like the cheetah who finds a different gazelle). Anyway, Player 1 sent most of his zealots to the mineral line of the main base and this time Player 2 had a different poor response: sending all his zealots to stop the attack. This left zero zealots defending the expansion and Player 1 cleverly kept a couple of his own zealots out of the attack so that they could attack the expansion uncontested.

Another example of this call and response was last week's game of Savior vs. Stork where Stork (Zerg) found early on that Savior was bad at responding to small harassment forces, so Stork's entire game was a series of small harassment moves all over the map.

Replay Analysis

Some students will be presenting a replay analysis as their final project, so Professor Feng presented a reply analysis himself to give an idea of what's acceptable. The match was Bisu vs. Savior (Protoss vs. Zerg). I'll keep it brief and just point out a few things. The big idea here is that Bisu is famous for his "Bisu build." If you are facing Bisu in a big tournament, what do you do? If you figure that Bisu won't use his signature build so you counter something else, then he DOES use it, you are going to look really stupid. Also, take into account that pro players like to be known for things, so Bisu probably really wants to use his signature build. At the very least, you should play in a way that prepares you to deal with it.

The Bisu build involves corsairs (air to air units) that take out the Zerg overlords (flying units that can detect invisible units), then dark templars (invisible units that attack uncontested if there are no detectors). Bisu did use the Bisu build. Savior did expect it. Savior's crazy play here was expanding to a base really far from his main, but right next door to his opponent's main. Why would he do this? The entire point of it was so he could build overlords there and get them into battle quicker! Floating overloards across the entire map or waiting for speed upgrade would be infeasible against dark templars.

Bisu's strategy actually goes according to plan until those sneaky overlords from next door come into play. I'll leave you with the match in its entirety: