BuzzerBeater Forums

Atleast for me it took two tries today. First it said "Failure, there may be too many connections or..." I waited two minutes or so, did it again and it went in.
rubbercube

My last training update isn't accepted by the website - it works great until i look at the history page and there is nothing new there...

I think it has to do with a player having a stamina drop on the same week- could that be the problem?

Is there a solution?

same here. i have send the training but in the last two weeks history has not updated. roster, synchronize etc works fine but the history does not work. any ideas how to solve the problem?

There's a new version of the GreaseMonkey on the site, install and restart your browser.

I would love to see the averages as presented in the training speed table, but separated for all combinations of age/height/coach level, in one big table. Like the raw data you give in the age table. If you ever have the time.

I'm having trouble visualizing this. You mean, instead of filtering (which I need to add trainer level to), show the data for every combination of: training type, skill, trainer level, height, age? I don't think there's going to be a good way to display that, especially since height is going to have a LOT of entries, many of which have a small amount of data.

If you just want the numbers, i think you can already get those by using the "Single pop data" link on the analysis page, which will show you every instance in which a certain training has been run, along with level, age, height, minutes and whether or not the specified skill popped. It's not averaged, but maybe you're just looking for the data?

Also, you might want to consider running new regressions for the skill weights and othe coefficients in the salary formula with the data you get in the week after the salary change. I'm afraid our current coefficients need an update (they were derived in the time before computers were invented).

Good call, i'll do this. Any reason to suspect the coefs themselves have changed and not just the scaling factors?

I know several people have been having trouble, and i don't have a great answer. For whatever reason, the host we use sometimes just times out. The perils of using a free hosting service, I think.

The best thing to do is make sure you're using the current script (it's not really much better at preventing errors, but it is much better at letting you know if there was a problem), and then if (when) you get one that fails, just wait a few minutes and send it again. I've had times where i've had to do this 4-5 times before it'll go through. Trust me I'm trying to improve it, but for now, just need some patience.

I don't think it has anything to do with the skill drops... there are plenty of those properly recorded in the database, and eventually I'll try to do something with them, but for now they're just silently being recorded.

The speeds you calculated are just the ratio of pops among all observations for this training type (and restrictions for age and height), right?

Yes. I've looked at the data a few other ways as well:
- the "get multi-pop data" place holder will be to display all training entries where we know the exact training history between consecutive--or several--pops in a given skill. i've tried using this to estimate how much each of those trainings contributed to the pop, but it was producing some funky data last time i looked. we have a lot more data now, so i should try this again.
- ultimately the best approach, i think, will be to assume a model and then fit all the single pop data to it. but, need to have a good sense of what the model should look like first.

unfortunately, both of these involve curve fitting, and not even necessarily linear curve fits. i really want to have this site as an ongoing resource, where the data just gets better and better over time, but that means being able to do the analyses on the fly. so if anyone has done any least squares fitting in php, i'd love to hear from you

Hm, I have a feeling it might pay off to look at this with fresh data. There may have been some small mistakes in the skill coefficients all along, for example highly skilled guards seem to be off their salary prediction a bit too often for my taste.

sounds good, i'll check after the update.