“I’m a GM top 100, and with my deep understanding of this game, like everyone I can say these changes are pointless and solve nothing. Instead of a secret balance council, wouldn’t it be better to have all game data from all servers and make decisions based on that data, like a smart company would? Plus, you have access to Google AlphaStar and DeepMind; they should create a system with a super-trained AI (with human APM) to play millions of games and use that data to determine what is imbalanced and what isn’t, making real balance changes instead of relying on this imbalance council.”
2 Likes
If you look at the AlphaStar data it has some pretty convincing evidence that Protoss is too strong. It was strongest with Protoss despite using the least APM and it was weakest with zerg despite using the most APM. That’s the same pattern we see in the human population inside Grandmaster. If you balanced based off of AI data then it’s going to say to nerf protoss.
I like the idea of incorporating AI as an additional data point but using it as the sole decider of balance would be a very bad idea mainly because those AIs are intellectual property of some other company, aren’t public, and weren’t designed for game balance.
Using data mining is very important to establish which race is advantaged and which is not and they definitely should do that not only for the pro scene but for grandmaster and the ladder too. But, if you did that, you’d also find Protoss once again overperforms. They are the most in Grandmaster for example. So, yeah. The only place Protoss seem to struggle is at the premier tournament level, and it’s doubtful that that is relevant to balance at all for a range of reasons. The skill differences between pro players are just too large and it’s that that decides tournament outcomes for the most part. If you were to balance the game around these skill differences, it would FUBAR the ladder balance beyond comprehension.
Wonder if they changed the apm caps when playing zerg. I mean you can get easily 100 apm more with zerg than with toss, depending on the style and situation. At least i get that. Another point is apm =/= apm. Spreading creep and building/morphing zerglings is not the same as blinking stalkers.
I actually dont want you to respond, i want to put it into perspective for other people which might not know that alphastar first got trained with toss, then they saw ooopsieee we had an apm cap on average, yes, but not in a certain situation so we could go ham on micro which completly ruined the approach. They fixed that later, yeah, but it still pulled off incredible micro. This is incredibly important that they messed this easy thing up so bad that it spoils their reputation to get zerg done properly. Because no doubt you will get worse results if the apm of zerg is similiar to toss.
For me alphastar was frustrating and a disappointment because the idea is so cool but the result was meh. An ai that did 10000 stupid things but was superior due to much better micro and macro.
3 Likes
Yes, they did. They made it mirror the human population as closely as they could. What it means it that there is a cost to having higher APM. Spending time doing actions takes time away from doing other things like thinking about your build orders. Your brain has limited things it can do and if it’s doing this then it’s not doing that. They stated something similar in their paper because they found that increasing the APM limits reduced performance in general. The higher the volume of actions, the lower the performance regardless of race, in other words. That’s a strong indication zerg’s apm requirements create the difference in performance.
That’s why the argument that “my apm goes up when I play zerg” is absurd, because you aren’t measuring how that increased APM load affects your performance. Even if you did, the effect is probably immeasurable on an individual level & only shows up in the average of a thousand players or more. If you aggregated such data, you’d have the league vs APM chart that I’ve shared many times.
Lie.
Show me the data.
Another lie.
Show me the video or article about what you are saying.
The data is publicly available in the AlphaStar paper:
AlphaStar is given the most APM for Zerg, and performs worst with Zerg and best with Protoss, mirroring Grandmaster and professional play (ESL cups) in both APM trends & racial strength.
The DeepMind researchers affirm that increasing APM does hurt performance. They offer a theory: doing tasks distracts from the grander strategy of the game, which is very similar to what I said. If you are doing tasks in the game, those tasks occupy space in your brain while doing them, detracting from your ability to focus on other things.
Funny enough, if you use the MMR values for AlphaStar’s protoss and Zerg, AlphaStar’s Protoss would have a 68.5% win-rate vs its Zerg which is a 2.2:1 win-rate ratio in favor of Protoss. So as you can see, AlphaStar’s performance almost perfectly mirrors Grandmaster (1.7:1) which mirrors pro play (2:1).
So yes. Protoss needs nerfs and the APM requirements are a strong indication that that is the case. Having higher APM is harder than lower, which means players have no reason to exert higher APM unless it is absolutely necessary. Zergs in Grandmaster/proplay simply offset Protoss’ balance advantage by playing faster than the protoss. The spending skill shows terrans offset the same balance advantage by having better spending skill, which is equivalent to saying that Protoss’ armies are smaller because Protoss suck at spending their money, and it’s this advantage that Terrans leverage for wins.
The AlphaStar paper is a good indication that skill metrics like APM and Spending Skill are reliable indicators of balance which means it’s easy to measure balance for the whole ladder by simply looking at APM metrics. You guessed it, the same trends observed in Pro/GM exist everywhere else on the ladder. You can understand why Protoss are having a melt down over the new patch. They know they are advantaged and are afraid the gravy train is coming to a halt so, naturally, they are having highly public meldowns across social media.
The new patch really isn’t that bad for Protoss, either, but the toss players know they have a big advantage and so they have confirmation bias that big nerfs are coming. This causes them to see the current patch as an enormous nerf to Protoss when it’s more like a 2% power reduction. The way they act confirms they know Protoss is advantaged. It’s so obvious that protoss is busted that even YouTuber’s like PiG are saying “Yeah, it’s busted, so what? Impossible to fix, move on.” He also wants people like me to stop talking about the APM differences because it underscores the balance issues he thinks are impossible to fix. But he’s wrong. It’s totally possible to fix this and we’re going to keep talking about it as long as it’s an issue. There’s no reason not to. I mean, why would we stop trying to improve SC2?
Oh, another fun thing AlphaStar provides us with is the proportionality of APM to win-rate. For every 12% EPM difference it creates a 2.2:1 win-rate skew.
You can´t read and understand what graphics show. Your analysis is just wrong. You are completely biased that can´t even see the truth right in front of your nose. Protoss just doesn´t win tournaments anymore.
Literally every data point shows protoss is advantaged excluding the semi finals of premier tournaments. Premier tournaments are simply outliers that do not follow the balance trends. This is data analysis 101. You’d encounter it taking a class on 5th grade level math. There are 10 year olds who know what an outlier is but somehow every protoss playing sc2 doesn’t know. What does that tell you about these players.
Literally. I just looked it up. They teach it in 7th grade math. 12 year olds know how to use the IQR test to exclude outliers from a set, but protoss players don’t. If you applied an IQR test to the finals of tournaments, you’d very easily conclude that those are not just outliers, but the outliers of outliers. They don’t follow trends by definition. When studying balance trends, they should be excluded from the data set. Why do protoss players keep using arguments that a 12 year old could beat?
The reason for this was blink, you can micro the stalkers perfectly, always blinking right before they take hull damage and continue to fight while a new shielded unit walks into place.
Just watch some of the PvP games and its pretty clearly demonstrated.
Same could be said for roach burrow micro. You burrow roaches and move them to the back. They heal while burrowed. It’s the same thing. Why did it have higher performance with stalker micro?
Not sure tbh, maybe: instant animation instead of burrow and unburrow, also stalkers can blink then continue to fight, roaches cannot they ahve to burrow move to the back unburrow to continue to fight. Gotta remember these are like 50-60 army supply armies too not late game.
Stalkers are also a bit tankier
Could it be that there are more roaches to micro and with apm constraints you get better results with fewer units.
They both cost the same supply, but im referring to uncapped games.
In uncapped ai games stalkers are the most valuable unit for micro it seems though im sure if it learned to immortal drop from a warp prism thatd be pretty good too
No in uncapped games the performance went down. Reducing the apm increased the performance. At those caps, it did better with stalker micro than roach micro. Why is that. The micro is very similar. One of the notable differences is that there are more roaches to micro. It’s a very natural explanation that roaches are harder to micro because there are more of them, and that would require more apm, when they already established that increasing apm decreases performance.
In order to do the same quality of micro, with roach burrow, you would have to do more burrow maneuvers, but that requires higher apm and higher apm is already associated with reduced performance. That’s why performance goes down even though the micro is practically identical. Doing more micro is harder than doing less. More units means more micro. AlphaStar does best with Protoss because it has fewer units, this reduces the amount of micro needed, that increases the quality of the micro and that increases performance. It’s that simple.
The same trends show up in the human population. It’s a fundamental trait of protoss.
Roach ‘blink’ comes with the downside that when the Roach is in the burrowing part, it’s basically useless. A stalker simply blinks back slightly but is still able to fight.
Roach burrow + pull back, while approximating stalker blinking to some extent, is just straight inferior. So I imagine if you give an AI the ability to do both, blink will always come out more effective.
Stalkers have longer range and more damage output than roaches do. This could frankly come down to a unit stat thing more than anything.
Also this.
Stalkers will bumble around in the back, doing nothing, the same as a roach. You want it that way because they regen health in the mean time. Roach burrow is actually stronger because you can do an in-place blink to reset the targeting without actually moving the roach, meaning it’s more versatile.
AlphaStar maximizes winrate. If roach burrow increased its winrate, it would do it. It doesn’t do it, and that’s because it doesn’t increase its winrate. The only question that remains is why doesn’t roach burrow increase win-rate when in theory it should. The obvious answer is that there are too many roaches to individually micro backwards.
By the way, the same exact thing happens when the stalker ball gets large. The players, and alphastar, stop blinking individuals stalkers and the strength of blink goes down. We know that even blinkstalkers follow this same rule, but, roaches don’t because sc2 forum players can’t admit the blindingly obvious truth that it’s easier to micro fewer units than it is to micro more units.
There is a process happening in your brain to identify the stalker that needs to blink, to move your hand, and to blink it. This process takes time. As long as that time is non-zero, that time limitation will put a limit to how many stalkers you can micro in a battle. AlphaStar will obey the same rule because it has a non zero time process as well and that means it has a limitation on how many units it can micro. It blinks stalkers but it doesn’t blink roaches, and that tells us the apm requirements of zerg micro are too high to use in a practical scenario. That’s blindingly obvious because you can’t micro individual zerglings even though in theory it would increase their efficiency.
Someone did this at one point in time. 100 lings can kill only 2 siege tanks with amove micro but with split micro all the tanks die and most of the zerglings survive. This is what roach burrow micro would do if it were possible to do it on the APM budget they gave the zerg alphastar.
One of the big advantages that Protoss has is that they have fewer units to manage and this is definitely, without a doubt, a big advantage. The number of units in a typical protoss army is low enough that you can do micro on it in practical scenarios with the APM budget of a typical grandmaster. The same isn’t true for zerg. To micro the same number of roaches, you’d need about 2x the APM. That means the process in your brain has to be able to replicate the same micro with half the time. It’s obvious why that’s impossible.