Blizzard did a good job, and further balance changes would mess the perfect balance of the current patch.
What change would you make for the game to be better? (or worse, we’ll see) Or maybe you too, feel that the game is totally balanced and should stay the same balance wise regardless of the acquisition by microsoft etc.
This is an age old debate . Is the goal of balance patches more about “perfect” or near perfect balance. Or is the goal of balance patches spicing up a stale meta game and making underused units more viable?
Talk about a super obvious troll post. Here are the current balance statistics:
Team 1 %
Team 2 %
These differ from last year in that PvT slightly favors Terran. PvZ and TvZ are basically unchanged. It’s also important to note HOW the balance ought to be achieved. Technically, rolling a die is totally balanced. Should the game automatically assign a winner by pure chance at the start? Of course not. The game should be balanced through a wide variety of viable strategies for each race, and the outcome of the game should strongly favor the better player.
Zergs have been playing ultra late game for years, and doing mass baneling in both ZvP and ZvT. Tier 1 and Tier 2 plays are basically nonviable for Zerg in any form. 1-3 base plays are also non-viable. Zergs only option is to hunker down, expand, and aim for hive tech. This is the polar opposite of a wide variety of viable strategies. The outcome of the games is usually highly random – usually boiling down to a multitasking mistake which by definition occurs randomly. Win-rates are heavily skewed in Terran’s and Protoss’ favor.
You’re doing it the riot way I guess. Can’t say I agree. It can be so many factors influencing those stats. Also, it’s not “random”. You make too many or one mistake and you can lose. It’s your personal responsibility in the end. You’re better or you’re not.
Mark it on the calendar. We are making a mistake this Tuesday at 8 AM during game 3 of our ladder session.
Nope. Ignorance of basic statistical principles is not a valid argument. The central limit theorem states that random processes converge on a bell curve regardless of their natural distribution when properly sampled and normalized.
I do generally like your posts, and I think the win-rate thing makes sense at a certain level. I don’t even want to argue, because you are a very smart individual. When Riot went to balancing purely off win-rate, I considered it a little ‘cheap’. Its not a bad way to balance, but it strikes me as un-ideal, and missing out on things, that only seasoned players would understand. You are a seasoned player yourself, and again it feels kind of weak to argue against your logic. It just feels a little weird to me to go solely off statistics, without deeply studying the game.
Troll strats, easy things should have a lower win rate. That can only be moderated by smart individuals. The numbers will lie there. Like if you push terran up when every terran is banshee rushing, it doesn’t work. The numbers will lie.
Here’s another one *snooze
Say 50% of protoss cannon rush
They lose every game
Well now, Protoss has 0% win rate
If you didn’t watch the games, well…
This is one thing that people don’t understand and why these statistics that they post are pure and utter SH*T.
Most of Protoss games involve a cheese in some way in order to win. Cannon rushing being the most prevalent (aside with standard proxy cheeses).
Your average player can NOT stop the cheese and loses causing the win statistic to be higher than what it really is. If you play the game normally you will see that Protoss fall behind HARD which is why the cheeses were invented and instead of balancing the game properly, the blizzard balance team in all their glory decided to just leave it in and this is what you have now.
Well, I mean, I did give two other criteria other than win-rate. One, there should be a broad range of strategic diversity for all three races. Two, the outcome should not be random but, rather, decided by skill (primarily).
Right now two players spam APM at each other for 20 minutes, and it’s just a matter of luck when one of them happens to make a multitasking mistake of some kind. Strategy is almost irrelevant; all strategies / tactics are picked for their ability to maximize the multitasking “pains” of your opponent, which allows a better multitasker to accrue advantages that wouldn’t exist naturally in the game otherwise. In other words, no weaknesses exist in the game other than your opponent’s APM value. With no weaknesses, there is nothing to exploit, and therefore there is no strategy. Trade-offs create strategy because you have to give up something in exchange for something else, which creates strong asymmetries in the game state. The goal of strategy is to find an asymmetry that gives you an advantage, and then leveraging that advantage to win.
Because there are no weaknesses (blizzard deleted them when they went on a crusade against “game ending moments”), there is no strategy except to abuse better multitasking and when you have two players spamming APM at each other it’s just pure luck who makes the first mistake.
So Blizzard really ought to redesign the game with a focus on strong asymmetrical game play, adding large trade-offs, and minimizing the role that multitasking plays. They should do this in addition to making win-rates balanced. This is difficult because with asymmetrical play one team will always be favored, which is why it’s important to do seasonal variations.
The map pool is supposed to do this, but in the current form it does not. They should bump up the size of the map pool to 50 maps, for example, and should drastically increase the diversity design of those maps. For example, they should abandon the “symmetrical” map design. They should have maps that are not symmetrical left/right or top/bottom nor diagonally. If a bad map gets into the pool, no big deal, just veto it and you’ve got 49 other maps to play on.
Frankly, if I were running the show at Blizzard I’d program an AI to generate maps. The AI would receive feedback about the maps in the form of replay statistics and players could also rank the maps from 0 stars to 5 stars. The AI would use probably a genetic solver with a high mutation rate and would seek to maximize a scoring function which took the various relevant statistics and combined them with weights. If it makes bad maps then people can veto them. But, it would practically guarantee some truly unique map designs every single season.
Also the way you evaluate strategic diversity using mass data mining of replays would be through markov chains. Evaluating strategic diversity is difficult because the game can be complex and it’s hard to properly evaluate all the possible win/loss conditions and that’s not even speaking to the difficulty of trying to attribute cause/effect to them.
So what you do is you take various game statistics such as unit counts at various stages of the game and you plot it on a frequency chart. You do this for all units and for all maps and all matchups. This produces hundreds of charts. From there, you generate markov chains that have “event” nodes which produce events any time the markov chain is on that node. An event could be labelled “marine created” or “marine died” etc. Then you generate billions of these chains and simulate them and plot their outputs on a frequency chart just like you did with the replay stats. The chains which most closely approximate the frequency distribution of the replays is the “correct” one. Maps which require more complex markov chains to properly model them are considered more strategically diverse.
Basically all a markov chain is is a series of links connecting to other links and these connections have probabilities. The simulation starts out at the start link, picks the next link at random according to the probabilities of the connections, and moves to that link, and repeats thousands of times. Some links just connect to other links in the tree, but other links are special “tagged” links which denote an event like a marine being created. This allows you to create a complex chain that outputs event tags at the same time/frequency as what happens in the replays. Since the timing/frequency of those events in the game is quite complicated, it takes a very complex chain to create the proper frequency/ordering/timing of events.
These systems are used to model all sorts of random processes ranging from the stock market to the weather and data compression.
With a way to evaluate strategic diversity from mass replay mining, an AI could generate maps that maximize the strategic diversity of the games. Maps would also be scored by combining, through weights, player feedback of maps, win-rates, and also a diversity scoring function which guarantees the maps are different from other maps. Since this would be completely automated, Blizzard would only have to shell out the cash to develop the system at the start, and after that point it would run on its own with no upkeep cost. The system would crank out bad maps at first, but after only a few iterations would start cranking out loads of insanely good and unique maps.
If you want to see a very simple example of a similar algorithm at work, check out this, which designs cars:
Now people read a post like this and think it’s too complex / “pie in the sky” / think "Would that really work? and “How practical is that!”. The answer is simple: yes. I’ve been doing essentially this for years in a variety of contexts ranging from bots for MMORPG games to engineering applications.