The future of rating systems after the Great Rebalance

[woe2you 367]

42 minutes ago, stagnate said:

 if the nerds are substantial

Not a nice way to be talking in Mathematics Corner...

[RichardNixon 835]

8 hours ago, erickHV said:

But is it possible to create a rating system that is completely impervious to various rebalances? Meaning it would assign players within a given percentile or rank or rating value based on their performance compared to their peers in a given patch, and then carry it forward "as-is"?

Bear in mind that the statistical effect of nerfs and buffs is unknown until they've been in place for several weeks, so any method has to apply retrospectively.

It's possible to create a database of nerfs & buffs together with their times (I already have one for significant 8.6+ nerfs), but it's not reasonable to enforce that in a rating system. Every site has access to different historical data.

 

6 hours ago, erickHV said:

Technically, how hard would it be to take a snapshot of all tracked player's stats at the end of the patch, analyze them and only store the final values of per tank WNx for that patch?

Assuming that your database management doesn't suck (big assumption), the main update cost is pulling all the data from WG's public API at a rate of ~15 calls per second. If everyone tries to do that on patch day then the API servers fall over.

 

5 hours ago, NThirtyTwo said:

In other words, keep a baseline log of the stats that matter and calculate the current stats based on the differences between current vs. log and proportionally update the log as required.

This doesn't work well with WN8 because it's nowhere near linear. In theory you could do it with WN9, although expected value updates would require an additional data pull for every account. You'd also lose the ability to display tanks played during an interval, and the ability to switch to a different metric.

[erickHV 4]

I have to say 15 calls per second (we are talking one account per call, correct?) seems completely insufficient. Maybe somebody should raise this topic with community contributors who have better access to WGs dev's and community people? It's not that hard to set up a stats DB mirror in a cloud server that's synced once per day or smth, this simple move would boost the call throughput enormously.

If the WG API server limits calls to 15 per connection (as opposed to global limit), you could even think about setting up a cluster of tens of cloud-based servers that would each stream API calls and forward them to a unified mirror server that ALL 3rd party stats tracker sites could access much quicker than now.

Other than that, if the game is going to go through a full rebalance, and we still want historical performance to have its place, I don't think there's any other way to achieve this but to create separate rating calculations for pre-global-rebalance and post-global-rebalance, at the very least.

[RichardNixon 835]

17 hours ago, erickHV said:

I have to say 15 calls per second (we are talking one account per call, correct?)

100 accounts per call to account/info (overall stats) and account/tanks (battles per tank), 1 account per call to tanks/stats (everything else). In theory you can get your limit raised, but I haven't tried. IIRC it's limited by IP so you can cloud-spam if you really want to.

Running a general proxy server would be very expensive (I'd estimate current demand at ~1gbit/s, with each snapshot at 40GB per battle type), although you could maybe run a specific "WNx server" at low cost.

 

17 hours ago, erickHV said:

Maybe somebody should raise this topic with community contributors who have better access to WGs dev's and community people?

I've seen no indication that they exist on EU or NA, but feel free to try. I gave up trying to deal with WG API devs. They clearly have their own priorities.