UC Berkeley StarCraft Class, Week 6
Saturday, March 7, 2009 at 1:11AM I think I'll skip the math this time because I can explain things just as well without it for this week in particular. Last week we talked about macro (building lots of units), and this week we looked at macro from another angle. The idea is to look at the entire process of manufacturing units to look at the bottlenecks.
Professor Feng emphasized that the entire point of good macro is to create units you can USE. Units that sit around not attacking, don't help. Piles and piles of minerals in your bank don't help. These things are only potential, but we have to cash in the potential to create an actual effect. Here's the six-step process of macroing up some useful units:
- Get income (minerals and gas)
- Make buildings that can produce units
- Make sure food/supply is high enough (supply depot / pylon / overlords)
- Produce units
- Get units out of your base
- Actually USE the units in battle
Think of this like an assembly line where a bottleneck at any point will shut down the line. Also note that dropping the ball at a certain point does NOT negate the steps before it, but does shut down the steps after it. If you forget to build pylons at step 3, for example, everything after that is stalled but any work you did on building your income is still valid. It's just that that work stockpiles in the form of "potential" that you'll have to cash in at some later time once you fix your assembly line (build the pylons).
Rolling Dice
Professor Feng also asked us to think about how at each step, we can imagine some theoretical "perfect" execution on our part, but we can also imagine making a lot of mistakes. For example, you can imagine building every single probe at the first possible moment, never putting any in a build queue, never missing a probe. But you can also imagine forgetting to build a probe for a few seconds here and there, or wasting some minerals queuing them, etc. So there's some range of possible executions here when you take human error into account. Remember that EVERY step has this range.
So how do these errors (your non-perfect execution) affect the assembly line? As a demonstration to help our intuition, Professor Feng took 6 volunteers from the audience and gave each one a six sided die. Think of the range from bad to perfect execution of each step as being 1 to 6. So we'll roll dice to simulate how well or poor we do at each step and see what happens. We put several rounds of this simulation on the board, looking at the effect. Rather than reproduce all that, I'll just tell you the gist.
It's immediately obvious that rolling a 6 on the later steps is a waste for the first few turns. For example, if you have only enough pylons to make a 2 units, but perfect execution of building units would let you make 5 units....that doesn't matter. You can only make 2 units anyway, so your amazing execution at building units out of your gateway fast doesn't matter.
Next, we saw up close what we should already have the intuition to understand: the early steps matter A LOT. Their effects ripple forward in time, so getting a poor roll on turn 1 income is devastating. That bottleneck ripples forward and slows down each future step at some point.
Finally, we notice that on later turns of the simulation, rolling high numbers on early turn stuff is a bit of a waste. Of the three effects I'm talking about here, this one is definitely the least bad though. If the assembly line is mostly to the point of moving units out of our base, for example, rolling a 6 on income doesn't do much. Often, we already had more than enough income anyway, and something later was bottlenecking us. But at least we can stockpile this income and have some hope of gaining a later advantage.
Remember that rolling high on later steps before we were really ready had NO help, and rolling poorly on early turns had a horrible, crippling effect. So the point here is to see which mistakes hurt the worst and which hurt the least. Professor Feng then asked us how we can do better, knowing all this. The answer is to focus your attention on the things that matter most. So on early turns of the simulation, income is HUGELY important to roll 6s on, followed by getting your production buildings up as efficiently as possible. We simulated this by giving extra dice to the first couple volunteers. This represents spending extra attention in real life on those tasks in the early game. As the simulation went on, we passed the extra dice down the line. Once we're at the later steps, having extra income isn't a total waste (it can stockpile at least) but it doesn't really help compared to removing actual bottlenecks at the later steps.
The takeaway here is just to be very conscious of where your bottleneck is at any given point of your macroing and make sure you're spending your attention THERE and not somewhere else. And if you have extra attention to spare (somehow??) after removing bottlenecks, you'd rather spend it on earlier steps that can stockpile.
Protoss Macro
Then we watched a reply of the player Tempest playing Protoss against a Terran opponent. This was a really incredible example of macro play in action. For several minutes, he built probes in the most efficient way possible, almost never missing the timing of a click. He build two gateways right on schedule, and pylons on schedule also. No bottlenecking at any of those steps. His gateways seemed to never sit idle even as he build more and more and more. In just a few minutes he was producing from NINE gateways with none of them idle and practically no minerals unspent at any given time.
Tempest sent huge armies at the Terran player, spending almost zero time on micro. Just attack-move and forget. Even though he'd lose his entire army this way, he'd take out most of the enemy army too. Most importantly, because Tempest had all 9 gateways operating at max efficiency, he had an entirely new army just moments later. There was just no way the Terran could keep up. I think Tempest got to 200/200 food three times in 15 minutes or something.
Walls
Most of the rest of the class was about the advantages of building walls. Terran are best able to build walls because their supply depots are so large that they can "full wall" meaning that not even a marine or zergling can get by. Building walls like this can prevent the enemy from getting into your base, which allows you to keep them in the dark about what you're building. It can allow you to go full economy with very few units or defenses, if you can hold your wall long enough.
Even a single marine behind a full wall can stop some zerglings or zealots from breaking through. Vultures trying to break through with pitiful concussive damage will take forever. If there's a real threat to your wall, you can send SCVs to repair the building as they are attacked to delay the opponent even longer. We also learned that if you have the choice to build, say, three supply depots vertically, that you might instead want to stagger them diagonally to make your wall. If you build them vertically, you can usually only repair with 4 SCVs at best on any given depot. If you build diagonally though, you can repair with 5 or 6 at once.
Even when you CAN make a full wall, you don't always want to. You might intentionally leave a small gap so small, single-file units can pass by. This is still a very good defense, because single file enemy units coming in are going to get hammered by just about anything you have set up behind. Imagine your horizontal line of 6 marines behind your wall, killing an infinite number of single file enemy marines coming through. It's still solid. But leaving a gap lets you move units out of your base when you need to. Also, once you're pretty far into the game, it's entirely possible that the wall (full or partial) is just getting in your way and no longer providing a real benefit. Don't be afraid to destroy your depots in this case. They've already served their purpose. (I think in StarCraft 2, Terrans can submerge their depots to let units pass by when needed.)
Walls don't need to be as extreme as these Terran walls to make a difference though. In this next example, the Protoss player builds a pylon behind the enemy's minerals, then a second pylon and a cannon. The second pylon at the top of the minerals is mostly to partially block the path to the cannon. Any drones that would try to go around the pylon to get to the cannon would be so slow as to die. Professor Feng said this tactic was initially very strong, but then players learned they could carefully move their drones between the mineral patches so they didn't have to go around the pylon. Here's the vid:
Notice that this kind of pylon/cannon placement is enough to make Korean fan girls scream repeatedly. You can also see one cover her mouth in amazement in that vid.
Next we saw a vid where a Terran player walled in the OPPONENT'S base, in a game where the opponent built his unit-producing buildings outside his main base. Ha! So the Terran player's few units inside the opponent's base had free reign.
Next was a pretty interesting replay of Protoss vs. Zerg. The map had pretty open terrain. The Protoss player expanded to his natural expansion, and build his pylongs, forge, and two cannons in such a way that blocked as much of the travel space as possible. It wasn't nearly a full wall, but it did mean that any enemy units would have their flux decreased and they'd have a hard time getting past the cannons. At this point, the zerg attacked with hydralisks, did a bit of damage, but was pretty much stopped by this configuration. He had to back off because if any hydras got in range to shoot the cannons, both cannons would fire back. Soon the zerg player had more hydras, but the protoss player then had 4 cannons, blocking even more area. He built them in such a way that 3 or 4 would fire on any Hydras close enough to attack any of the cannons. Very good.
The Zerg player sacrificed his economy to do a 2-hatchery, all hydra rush with upgrades on hydra range and speed. Meanwhile, the Protoss player had almost no units at all. He went nearly full economy at two bases and used this semi-wall to hold off the attack. Even though I thought that was pretty good playing on the Protoss player's part, this is considered Zerg's advantage at this point. Yes, Protoss has a very strong economy at these two bases while the Zerg player has a horrible economy. But the Protoss was contained to just these two bases while the Zerg was then free to take over the map and switch to full economy. 30 minutes later, the Protoss player went on to win, because he is generally awesome from what I gather.
The last example of walling was pretty awesome. A zerg player used his hatchery to kind of block a choke point, but not totally. The enemy then sent some zerglings or something in, and just as the enemy units were getting close, the clever Zerg player build three drones. The three eggs that appeared from the larva BLOCKED the path and created a true wall! Ha!
When one student asked how this could ever be repeated because the movement of the larva is pretty random, the professor said that you can select the three larva along with an overlord, then use the "stop" command to exploit a bug that makes the larva move to the lower left. Crazy.
Final Project
Professor Feng also reminded the students they will have to submit a final project, 2 or 3 pages long, that analyzes some aspect of the game that isn't already analyzed. For example, a function C(t) that shows the damage over time that carriers do in a certain configuration. That's actually really, really complicated. He advised the students to start with ideal conditions then add more and more game elements to the calculation to make it more and more realistic. In addition to this, they also have to submit an analysis of a replay of a pro game.
They also have an alternate option instead of all that. They can challenge Yosh, one of the class facilitators (and 10-year tournament veteran) to a match. Win and you pass. Lose and you fail. At first it seemed that beating him meant only 1 game, but then it became best 2 out of 3 games...which is even harder. (If your opponent is better than you, you would prefer as few games as possible to hope to get lucky. If you are better than your opponent, you would prefer as many games as possible in the set, to factor out any random flukes.) I hope someone initiates this challenge for the sheer drama of it, but I doubt anyone will. ;)
Reader Comments (14)
Hmm, can you elaborate on the PvZ match? When Zerg goes for a hydra break, getting mass hydras to try to break Protoss's cannons, it's usually a win or lose moment, since Zerg has to sacrifice a lot of economy to produce those units. If he's just trying to contain Protoss, he'll usually just get lurkers and a few lings instead, while he expands all over the map.
I'm a little confused over that game. Every time I've seen mass hydras it's been a timing attack before high templar can get psi storm.
Response by Sirlin: The Zerg player's goal wasn't to contain with the hydras. His goal was to just win with them, as you said. Yes he sacrificed his economy, as you said. It just ended up that the Protoss player was able to survive it, but surviving left him with almost no units and no way to stop Zerg from taking the map. Sure it would make sense to use lurkers to contain if that was your goal in the first place.
Bah. The problem with a challenge like that is that from a PTW standpoint the risk/reward is WAY out of whack. The only ones who would ever be willing to take that challenge are those guaranteed to fail any other way.
... unless maybe they had a way to cheat? Pay the guy to lose, or have a friend screw with him on remote desktop or something. If they paid attention in class, they'd know the best times to screw with him too.
Even as someone who never played starcraft, I'm sure I'd have loved the opportunity to take a class like this.
Do you think in SC2 it will become a more acceptable practice to change your hotkeys? I've posted on starcraft league forums about this and I basically got called out for being extremely noobish and that changing the hotkeys would somehow undermine the skill requirement of gameplay.
Response by Sirlin: It seems like you know what I would say. You can balance a game like this:
1) make any UI you want
2) balance the game around it
But it's also possible to balance a game like this:
1) make the best UI you can, including sensible hotkeys
2) balance the game around that
Approach 2 seems much better because you can end up with balance either way, and you might as well have non-terrible UI. It sure would be nice if every you could make every spellcasting unit use the same button for its primary spell, for example, and the same other button for its secondary spell.
lol, Yosh is one of the top 10 USA players.
Because of the need to optimize the first few steps of the game, do these steps end being the same every time?
I mean if you execute perfectly like that Protoss player, then that means the very beginning of the game is devoid of any choices. I'm sure he could have built less gateways, but before that? Isn't that boring?
Response by Sirlin: It seems that these steps don't have ZERO decisions, but they do have very few. There was basically one decision (economy + 2 initial gateways) that governed the first several hundred clicks. The rest was an optimization problem unrelated to strategy and unrelated to 2-player interaction. Of course more decisions happen later, but those first few hundred clicks do seem more like Guitar Hero than a competitive game. For some reason this is looked up to, rather than looked down on.
I think it would be more interesting if you had to do a make-up assignment or fail if you lost to Yosh, since then you'd probably get a contender or two. At the moment, the possibility of failing seems too strong of a deterrent.
I know WC3 allows you to remap hot keys so I'd hope they'd carry that forward into SC2.
Challenges like that in classes sound like a fun idea, but they're always so awkward for everyone... how bad do you feel if you're Yosh and you're going to make some guy fail who had the guts to challenge you? I feel like a way better system would be like, if you beat Yosh you don't have to do the match analysis, and if you lose, your paper has to be twice as long and in depth, or something like that.
Response by Sirlin: That reminds me of how in ST, T.Hawk is all or nothing in some matches. It's so hard for him to get in, and he needs his inescapable loop to auto-win once he does. But in HD Remix, the loops is merely good, not 100% inescapable, and it's a bit easier for him to get in. Not so all or nothing anymore, and better for it I think. Facing Yosh for the entire pass/fail is as all-or-nothing as you can get, ha. So yeah, I like your idea.
@ Marc.
Opening play in Starcraft is very similar to opening play in chess. After much study, there are simply a handful of openings that are better than other, and they will almost always be used. That said, the nuances of these openings are incredibly varied, and TOTALLY dependable on what your opponent is doing. Just because a casual observer is oblivious to the reasons why gates are added at certain times, why certain tech is added at certain times, etc, does not mean that it isn't a very conscious decision by the play, which is governed to a very strong degree by what his opponent is doing. Ask yourself, is the only player interaction in chess when a piece is captured? Of course not. Opening game is incredibly full of choices, it is just that unfamiliar viewers don't see them.
@ Sirlin. Of the first 1000 clicks in a game, 90% of it is spam. I know you want to think that some incredible finger gymnastics are necessary to pull of early game, but it is more akin to mine sweeper than guitar hero. The reason why people click so much in the start is same reason why the players, in your previous example of an attention limiter, would click on the gateways instead of just looking at them while they got the fast building bonus or w/e.
More big-picture-oriented RTS games have a novel solution to the "play Minesweeper for 1000 clicks then play the game" problem: deficit spending. Resources are deducted from your account gradually as each unit is built, allowing you to queue up several units if you know you'll be able to afford them down the line. There's still a small number of optimal opening moves, but they are easier to execute because you can use the queue function without wasting resources.
I wonder what pointless semantic argument is going to come up in this week's discussion!
But seriously, thanks for these class reports, Sirlin. They're very interesting to read.
Although the first 10000 clicks are mostly spam, I believe the reason they do it is to "warm up", because later on in the match more of those clicks will be meaningful. I still believe that there is far too much clicking to do menial tasks in Brood War. I look forward to what SC2 brings.
Yes, it's purely to warm up.
Everyone can try this for themselves... Play ANY RTS which requires ~150 apm to play really good.
Now at every point in the game ONLY do the required actions (this would mean probably ~10 apm per minute in the very beginning of Starcraft, but it accelerates pretty fast). It will be way harder to just jump to your maximum speed if you did nothing in the beginning. It's really warming up.
From the first ~5'000 clicks about 4XXX may be obsolete at this stage of the game, but later it helps you to keep your speed.
Btw: The other XXX clicks you make still take the game into a very distinct direction (like rush, pressure, economy, tech to keep this very basic...).
Response by Sirlin: I think many have missed the earlier poster's point. His point had nothing to do with all the extra warmup clicks. What he meant was if Tempest has decided to go with this build where he gets 2 gateways fairly early but never misses a probe build, never misses a pylon, and executes that build perfectly...then isn't he just enacting a script, of sorts? Isn't that beginning portion of the game filled with many, many clicks that aren't related to decision? There was really only one decision between the beginning of the game and the time the two gateways arrive. I think the answer is "yeah that's right, that's just how the game is."
Your right/wrong well...
It's like in chess or most other strategy games. As long as your opponent is not forcing you to do something, you do the thing you planned as perfect as possible.
You decide to go for Opening X and the next few turns or clicks you basically just do that opening, you could also write a script for these first few steps in Chess or every other game. At least if both players play "conservative/defensive"
If not for the possibilties of fast rushes and you yourself screwing up the Build Order you planned you could *script* it.
As long as no enemy is harassing you or/and you have no information about what your enemy is doing, you are following the build order/script you have chosen to go with beforehand as close as possible. Everything else would be pretty stupid, right?
As soon as you got scouting information you adjust your BO/Script if it would be beneficial for you.
I mean your Build Order with Protoss is kinda decided by the supply amount and the location you put your first pylon down and after that by the time you get your next building.
At the start of the game you just follow your chosen strategy/build order and because it is an RTS you have to do it as precise as possible.