MMR and Tanking?

Demiurge_Will

Psykopathic wrote:

Demiurge_Will wrote:

It's very odd to me to see a (I assume male) with the xxxxxx-xxxxx last name. Maybe it's a weird city people thing.

You caught us. This game is made by weird city people.

Demiurge_Will

Bugpop wrote:

Why does it seem that there could be as many as 30 opponents to choose from and at other times, as few at 6?

Every time we search for opponents, we guarantee that there will be at least 6 in the batch or we expand the search radius. There will almost always be more than that (especially with the recent changes), but if there are very few people close to you in hidden rating and high enough in event rating to be worthwhile for you, it's possible to have as few as 6 players graded a "best match" for you.

Unknown

Can something be done so that there's a guarantee of at least, say, 2 or 3 people that will get you at least 25 points, so that there's not just a bunch of net-negative people? With the current retaliation and scoring system, attacking anyone below 25 points means you're liable to lose points overall even if you win.

Unknown

Ben Grimm wrote:

Can something be done so that there's a guarantee of at least, say, 2 or 3 people that will get you at least 25 points, so that there's not just a bunch of net-negative people? With the current retaliation and scoring system, attacking anyone below 25 points means you're liable to lose points overall even if you win.

This has been my exact experience. I've been trying some late-night lightning rounds to maybe increase my 3* collection, and I find I reach a point in every round where: a) My only options for attacks are 15 points or less, and b) I lose 25-60 points in the time I'm able to play a single game (even when I'm managing sometimes a game in [very rough estimate] ~1).

I'm not ashamed to admit, I have now begun tanking. I think I've been doing it the old fashioned way for a while to some extent... that part I might be a little ashamed of...

Unknown

Demiurge_Will wrote:

We're making and have recently made changes to matchmaking & the rating system, a couple of which are aimed at making tanking less important, less appealing, or less effective. (One thing we did recently is shift the matchmaking search window to give more weight to your event rating relative to your hidden persistent rating.) I'm not sure that we can ever totally eliminate manipulating your rating to gain an advantage - it's an issue in every sport with a rating system, even ones without asynchronous play and shields and all the other things that make rating in our game weirder than it is in, say, chess. But yeah, we hear you, it's a quirky way to get ahead and it can be less effective than it is now.

We move slowly with this stuff because it's often very hard to predict what will happen with a rating/matchmaking system before you see it in action with a real player population, so we make small changes and watch for a bit before making another change, but you can expect more changes designed to make this better in the future.

Err, excuse me but WHY is it that hard to make predictions? I'm positive the tester guy sitting behind me could hack together a simulation system in a matter of hours if I was developing changes on the said system, and asked for outcome. You start with the realistic rate of attack success (my estimate is 95% first round, +4% second round 1% loss is a good start). Issue some selection strategy -- say 3rd best score from all presented, or first over 24 pts gain that shows up naturally. Then just run the simulation with play times randomized between 30-200 seconds.

It's not much work to make simulation for the old rules and match it against historic data either for checks.

With mere 1-2 man-days of effort a plenty of strategies could be added, as the game itself is not subject to analysis, being auto-win after the opponent picked.

Unknown

pasa_ wrote:

Demiurge_Will wrote:

We're making and have recently made changes to matchmaking & the rating system, a couple of which are aimed at making tanking less important, less appealing, or less effective. (One thing we did recently is shift the matchmaking search window to give more weight to your event rating relative to your hidden persistent rating.) I'm not sure that we can ever totally eliminate manipulating your rating to gain an advantage - it's an issue in every sport with a rating system, even ones without asynchronous play and shields and all the other things that make rating in our game weirder than it is in, say, chess. But yeah, we hear you, it's a quirky way to get ahead and it can be less effective than it is now.

We move slowly with this stuff because it's often very hard to predict what will happen with a rating/matchmaking system before you see it in action with a real player population, so we make small changes and watch for a bit before making another change, but you can expect more changes designed to make this better in the future.

Err, excuse me but WHY is it that hard to make predictions? I'm positive the tester guy sitting behind me could hack together a simulation system in a matter of hours if I was developing changes on the said system, and asked for outcome. You start with the realistic rate of attack success (my estimate is 95% first round, +4% second round 1% loss is a good start). Issue some selection strategy -- say 3rd best score from all presented, or first over 24 pts gain that shows up naturally. Then just run the simulation with play times randomized between 30-200 seconds.

It's not much work to make simulation for the old rules and match it against historic data either for checks.

With mere 1-2 man-days of effort a plenty of strategies could be added, as the game itself is not subject to analysis, being auto-win after the opponent picked.

It is always easier said than done.

I reckon to build a simulator for black box testing is not a one day work. It's more like 1month work.

AI could improve, but not without millions of trail and errors. Take chess for example.