Machine Learning to Balance Game at Every Elo

picotan·3/23/2017, 6:11:21 PM·1 votes·578 views

Dear Riot Games,

I am not very experienced as a developer--but it occurred to me that there might be a way for you to auto-balance every champion with a machine learning algorithm. All that you need is an environmental variable on the server side that provides micro buff/nerf for each champion at that particular elo. So one for bronze, one for silver, ect..

If a champion wins more than 50% of games then the micro nerf slowly increases. If they lose more than 50% of games the micro buff slowly increases.

Then you can make whatever changes to the champions you like and the game will automatically balance the champion at each Elo.

Cheers, Peter Davis

6 Comments

BlastRush3/23/2017, 6:16:14 PM1 votes

Are you trying to say that they would change champions based on ones elo?

Thrëat3/23/2017, 6:20:05 PM1 votes

There is a huge difference between the win rate of a champion and whether or not a champion is balanced. A champion can have a 40% win rate or a 60% win rate and still be balanced. It could be a champion that takes a lot of mechanical skill, so those that mastered it can do well while others that have not will sit at around 20-30% which will bring it down to 40%. There can be a champion that does really well at a certain elo because of an item, rather than the champion, being unbalanced, which could lead to a 60% win rate if that champion happens to be utilized a lot with that item. Also this would be a nightmare in higher elos, where for some champions you would be balancing based on a handful of players. Certain champions are rarely played in Master/Challenger, or even parts of Diamond. Which means that the entire balance for that champion could be based off of a small handful of players that actually play that champion.

Champions should not be balanced for each rank, they should be balanced across the board.

Zielmann3/23/2017, 6:21:12 PM1 votes

Theoretically possible if you limit 'balance' to be strictly numbers-based. Unfortunately, that's not reality at all. In many cases, issues are due to other aspects of a kit, items, or maybe masteries. Riot has given themselves so many levers to pull now (as they like to word it) that adjusting on the fly like that would be incredibly complex.

Also, the assumption that >50% win rate = too strong is shortsighted at best. Some champions might have an overall high win rate at a given ranking, but it's because they have a very low pick rate, indicating that only a dedicated few players are taking that champion in that role. Perfect example of this right now is Support Sion, as shown on champion.gg right now. He's the highest win rate support right now at 53%. This would lead to nerfs to Sion under this system. But the pick rate is a dismal 0.61%, and remember that this is Sion support. The nerfs would hit top lane Sion, too, who actually has a lower win rate than support Sion.

That's not to mention how there would then be no way of knowing/tracking a champion's numbers. And then the strengths of the champion would change as you climb in ranking which could potentially feel horrible.

Sasogwa3/23/2017, 6:21:41 PM1 votes

Riot's philosophy is to have the same game at every elo. They do not want to have a game that is different in numbers according to the elo (and to be fair, it would feel fucked up)

WoonStruck3/23/2017, 6:26:51 PM1 votes

Or they could just balance it for the top end.

If a bronze player thinks Garen is unfair, they can just get better at the game.

If a diamond player is abusing Ryze or whomever to win consistently against other good players, they can't just 'get worse' so that he feels balanced for the other players he's against.

If something is 'unfun' to play against, there could obviously be an argument to rework them. Riot's started doing this even when champions AREN'T considered 'unfun', so there's really no argument against it.

Micbran3/23/2017, 7:56:32 PM1 votes

Don't balance for low elo.