With the fluctuating time that Berig and I have, we have decided to adapt an Agile development cycle, where we will be pushing you new features often, and let you guys find the bugs. Pretty much, thats an excuse for not having enough time to complete the project in its entirety, but not wanting to keep it from the hands of our fans. We'll keep up on the features and fixes. Our top priority is the tracking, but with the user pages down theres not much we can do.
So here you go. StatAddict 4.0 Beta release. Enjoy.
http://www.StatAddict.com
Known Issues
You are probably aware of this, but there's still something wrong with the way you're doing powerrank estimates. It looks like you are dividing by the total number of users instead of the number of users between the two closest known spots.
On an unrelated note, I was just thinking: Because the powerlist updates automatically, but the user pages (and hence the known stats) update rarely, you can get some bizarre situations with your powerrank estimates in the top 1000 from my understanding of your algorithm. For example: Imagine AceoStar is 1m total stats and Rokugo is 1.1m total stats. Rokugo goes on vacation and the userpages don't update, but Aceo continues to hunt. When Rokugo comes back, AceoStar has become 1.2m total and Rokugo checks stataddict. Rokugo is now 102 on the powerlist and Aceo is 100. But since the userpages haven't updated, StatAddict 'knows' that Rokugo's stats are 1.1m and Aceo's are 1.0m, so when it tries to calculate AllyGator's stats, who is 101 on this hypothetical powerlist and doesn't show her hypothetical stats, it calculates her stats as 1.05m. So when the powerrank estimating algorithm goes through, it leaves something like this:
99. Frog 1.24m
100. AceoStar 1.0m
101. AllyGator 1.05m
102. Rokugo 1.1m
103. Pythagoreus 1.02m
(This is what your powerlevel estimation algorithm would see (I think. It should preserve top 1000 order when it combines the top 1000 list with the known stats list, right?), obviously when you do a sort on the powerrank levels, it'll reorder itself, but if there were two people between me and you, they would be in backwards order, and I would be above you on the powerlist)
So I'm curious what you're doing to rectify this, because it seems to be a fairly big problem considering how rarely the userpages can be updated. Right now it looks like you took anyone ranked too high and put them right below the person whose known stats are the next highest right before you do the powerlevel estimation. Example: Even though Sniddster is well above Genica on the powerlist, on stataddict he is ranked right below her since his month old stats are below hers. Also, Hydekun, who is between them on the official powerlist, is ranked right above both of them on stataddict.
Yeah, Im gonna have to leave those tweaks for another time. I never got around to putting them back in after our last discussion :/
Sad. Too bad they don't make a paypal button for donations in the form of free time.
Also, thinking more about the 2nd part from above, I've come to the conclusion that this behavior comes from the way you combine the top 1000 and the known stats people. I also can't think of any way to both preserve known top 1000 order and still give reasonable estimates 100% of the time, so I guess what it's doing right now is as good a way as any to do this.
Aw, I'm on the hypothetical bugged list, how sweet.
Right now, I store local copies of the user pages that I'm parsing to reduce the update time for development. I think i need to start caching the top list as well to fix the problem you mentioned Rokugo. Because you're right, once the userpage data is old but the top list isn't, you have a hard time combining the two lists together. I hate to force some data to be older than it should be, but I really don't have much of an option because of constant downtime on userpages.
I knew this was an issue, but I really appreciate you looking into the data as well and seeing it too.
Heres the code again for my estimate
You'll need a few more variables to work with, but once you have them, it's pretty simple. I recall you mentioned that adding the variables would be nontrivial but doable.
^You're the man. I've got this working now and Im just adding some extra tweaks. It is doing a great job of making estimates for the top 1000. I should have it running with the updates here soon. I have some ideas for fixing the lowbies a bit.
Good, almost nobody above me shows their stats, and I'm tired of doing it by hand!
In all seriousness, I'm glad to see SA back, and I hope that its renewed popularity will convince mug that fixing the userpages is something players are interested in. Thanks dude <3
Killa.
Got the new spread running with the updates. It sets winroute to 10m power, but Im sure he's not far from it, so I have no problem with it. See it all here. If you want to help, let me know how far off your estimate is if you don't show stats. Mines almost dead on
http://www.stataddict.com/report.php?&vita=&mana=&powerrank=&sort=powerRank
I really think if you want to increase accuratcy ten fold, that you need to leave winroute as an outlier and keep him from your approximations. He is grossly overstated to anyone else in the game.
I think you should take the two known points you have, and have your program guess up to aubrey.
The only powerLevels really affected are the top 10 or so. As long as each of them is a portion higher than the previous, that'll help with weighting issues.
If we switch down to estimating to aubrey, we'll still have to give her some sort of estimation to go on. How about you guys bug someone in the top 10 to show thier stats. Im pretty sure subchess showed them at one point. Ceez seems to be unreged but I think he's in top 10 also, so that may help.
I just wanted to report that when pulling up the menu for subpath chung ryong, Assult is spelled wrong and could be the reason the page doesn't show up at all.
Looking good! Welcome back!
Seems to take longer then it used to to view lists.
When do you think you'll have crafting "power" lists available?
^ - We're collecting all of that information now, so we should be expanding our lists shortly.
We've gotten some really good feed back on the estimates. Most people are finding that it is pretty close. Some of the top 20 have confirmed anywhere from a 100 powerlevel difference to a 1million.
I have a sa warrior that doesn't like to post his stats. However, if you promise to keep them confidential, I can provide them for you to help you reach your accuracy goals about stats.
Can you keep a secret??
Absolutely. We're currently doing that for about 5 top 20s
the entire community has less gold than amaroq ...
Powered by Invision Power Board (http://www.invisionboard.com)
© Invision Power Services (http://www.invisionpower.com)