Dear Valve,
It is no question that your match making service for TF2 is horrible. Players are met with long wait times, unbalanced matches, and matches the end as soon as they get there. Players, much like me, get infuriated about how i get matched onto a loosing team at towards the end of the round every time.
you need to change something about this process, Team Fortress Two is loosing players because of it.
Before the Matchmaking Update
Before the Matchmaking update, the "Valve Servers" were searchable servers that the client-slash-player could choose from. Servers would never stop, and players could set the round to be whatever map they wanted. Experiences players would join them to use the non experiences players as cannon fodder, and the non experiences players would join because they didn't know how to use the game yet.You most likely changed the way players join servers for this reason, so that newer players could go up against players that wouldn't decimate then instantly. However, you only focused on one aspect of match making, and left out all of the others.
Machine Learning
Most of the problems that are facing you right now are problems that are very hard to fix. Problems like team balancing and team composition are problems where one fix will lead to another problem. If class limits were in place, a player may not be able to play the class that he wants.
I am suggesting that you implement machine learning into the match making system. Yes, it will take a very long time for the TF Team's six coders to make machine learning for a match making system; but you could make the code available for all of their games that need match making, which will make it much more worthwhile.
Machine learning could make a perfectly optimal system; where The Machine does the best thing for everyone. Each client would report what it wants to do to the Matchmaking server, and the server would take into consideration skill, map preference, class selection so that there are not about 4 spies and snipers on one team, and the time that the player plays the game, in order for the Machine could know when the player is expected to log off, so a ton of players don't suddenly leave. The data would come from Steam Community, the servers themselves, and the clients.
Auto Balance
Machine learning has the power to code itself, and notice and consider the smallest of detailed that programmers would not otherwise take into consideration themselves. Team composition can make or break a match for a team. Machine learning can learn what players do in certain situations, and make a team perfect for a match, accounting for team composition and skill by using the profiles of the player.
Crisis
The Machine, in order to avoid something going horribly wrong, should have a period in which it only observes, and learns. Although, it will be very hard for a machine to learn how to change something for the better, if it cannot change it. The Machine, for it can learn at all, must be able to experiment on the matchmaking process; though there needs to be a safeguard
The safeguard does not have to be a complex system that has thousands of lines of code. The safeguard can have two simple rules; one: the average wait time cannot exceed 2 minutes, and two: no player can be waiting for more than 5 minutes. I just noticed that the simplicity of the given rules shows how bad the matchmaking system currently is; you can't really screw up TF2's matchmaking more than it is.
The Pyro Update
[Hopefully] there is an upcoming update that will add more weapons for Pyro, and add balance changes and other cosmetics, and maps to the game; so Valve, I am openly encouraging you to add machine learning to TF2's matchmaking process.
Sincerely, OneTwentyEight.
Comments
Post a Comment