RealGM

EvanZ wrote:

jman3134 wrote:http://basketballvalue.com/index.php Is this website so last year with the xRAPM out now? (and I say that in gest b/c they only have last year's numbers) Can someone (Evan) explain the difference in APM vs. xRAPM?

To the first question, the guy who ran basketball-value (Aaron Barzilai) now is the head stats guy for Philly. He probably won't be updating the site going forward (I spoke to him over the phone, and while he didn't rule it out, I would say he sounded very doubtful about it).

To your second question, it's really fairly complicated. In a nutshell, RAPM is the regularized form of APM. RAPM is a better predictor than APM. xRAPM tries to use box score stats to improve the predictability of RAPM. (Apparently) it is a better predictor than pure RAPM.

Do you know much about what's involved in that? Cus +/- stats of any kind are like the opposite of box score stats, it seems like adding in things like point, rebounds, assists (things where the exact value of each one isn't always...exact) doesn't seem to fit into the PM model.

-Kees- wrote:Do you know much about what's involved in that? Cus +/- stats of any kind are like the opposite of box score stats, it seems like adding in things like point, rebounds, assists (things where the exact value of each one isn't always...exact) doesn't seem to fit into the PM model.

In theory, any orthogonal data should be able to help predictive power (or at worst not hurt), including box score data (introduced to these models as priors). If you want to read a very technical discussion of how this works, read this article (which was doing something like xRAPM before Jerry did):

http://arxiv.org/abs/1301.3523

Here's the original APBR discussion:

http://apbr.org/metrics/viewtopic.php?f=2&t=247

Hell, I forgot to plug my own (new) site. See my sig. It's pretty cool.

I know this thread is mainly geared towards the NBA, but has anyone had any success acquiring game-by-game statistics of college basketball games for the 2012-13 season? Specifically, I'm looking for game stats that would allow me to attain my linear regression models for effective FG%, turnover %, offensive rebound %, free throw rate, assists rate, blocks rate, and steals rate. Via a few scripts doing some screen-scraping, I was able to acquire all game-by-game statistics for the 2011-12 NCAA season. I'd rather not go that route, however, and would obviously prefer a quick-and-easy file download if one existed. Any assistance would be appreciated. Thanks!

http://stats-for-the-nba.appspot.com/ no longer has current data either. It's getting bleak.

spam

Here's a great new Android\iOS app that lets you follow or create automatic sports stats streams from sports-websites so you can see stats evolution graphically over time and follow other user's statistics. It lets you build your own personalized stats board:
http://www.numbeez.com


Its under "Numbeez" search in the appstore\google play
It's still new (So the content is still limited), but it's realy cool and you can take part in creating some interesting stuff there
Enjoy!

Im very interest in ASP,RAMP,VORP and Hollingers EWA for all-time greats ( every season leaders, career numbers, single seasons perfomances), where I can get it not only from 2001, but from 80s at least?

Is there a way to take raw data from a non-nba league to come up with these analytical formulas?

I am gathering data for a classification model for NBA games. The model will predict the outcome of a NBA game, based on the given attributes. I am having a little trouble with the data.

I like the website basketball-reference, but I dont know how to retrieve the data. For example I would like the Box Score for every game in a season, like you see on these link http://www.basketball-reference.com/leagues/NBA_2013_games.html. Are there any websites that offer data in a format like csv available for download?

I will parse the html from basketball-reference somehow if there is no easyer way.

I contacted Stats Inc to see about the same information and the sales rep let me know that $3k per month is what I was looking at or $20,000-30,0000 if it was going to be distributed publicly in any manner. That was a couple years ago not sure if it changed.

If your purpose to predict outcomes of games you do not need all of that. Even if it was to predict a players points scored, unless it is matchup based but even matchup boxscores change so much in a small sample size it would be very hard to actually get something beyond an advance average metrics.

So what is your ultimate idea to accomplish?

I

My ultimate idea is a predictor of outcome of games. I am going the use the Basketball reference HTML, the parsing is not as bad as I imagined and I am not going to spend money for this data. Perhaps I don't need a boxscore. So far I have made a dataset which is composed of examples(one game) which look like this:

Orlando Magic,Miami Heat,0.6,0.2,0.9,True

away team, home team, percentage of home team victories in h2h in last 10, away team form, home team form, home wictory

Why I wanted the boxscore? I will add more attributes for the predictor. I was thinking that one attribute should represent the value of the injured players, so I would in the boxscore if a certain player played, though it is true that we may not know that before the game starts. Perfectly I would want the list of injured players which is announced before the game.

TheToothFairy wrote:Is there a way to take raw data from a non-nba league to come up with these analytical formulas?

Anyone?

TheToothFairy wrote:

TheToothFairy wrote:Is there a way to take raw data from a non-nba league to come up with these analytical formulas?

Anyone?

Probably. Why don't you invest the time trying to figure it out?

Sorry if this is not the best place for this...

But is 82games.com gone for good? It's been unavailable for several days, and was wondering if anyone knew if it was coming back or just shut down all together. Thanks.

http://stats.nba.com/?GameScope=Playoff ... ory=Points

STATS, LLC and NBA just teamed up, so this might be the go to source in the future, but only time will tell.

Also, which website is the best site to use for updated daily stats? Basically I'm looking for updated stats after all the games for the night have finished.

Any advice will be helpful. Thanks!


Sent from my iPhone using RealGM Forums

I was told I need to post some over here at RealGM - so I'm posting here first, since I'm an analytics guy with an analytics site (and since it's November, sportin' an analytics 'stache).

http://www.hoopsnerd.com

I'm doing complete NBA player rankings every day according to my NBA WAR (Wins Above Replacement). I'll update my All-WAR team (top WAR/minute guys) every day.

Don't mind the site asthetics - I have no concept of form/funtion with websites. It'll get better in time.

I also do college basketball player ratings/rankings. All D1 player rankings (SoS, pace, PT relation to teamates are all accounted for) over the last 4 years are at the site. I'll be doing complete D1 player rankings probably once a week come sometime in January (when the team ratings "noise" settles some).

So, anyone interested in seeing complete player rankings based on my form of analytics - check it out. I'm @Hoops_Nerd on twitter if you wanna follow my updates or just yell at me.

Hoopsnerd:

You need to go to Basketball Reference and download past seasons and work on your WAR. Almost any analytics will show players like Lebron at the top. The key is when you truly see the difference in players near the bottom. Sometimes they need to be high profile because the media and fans tend to hype players that absolutely are a detriment to their teams when you are talking an average of 4pts difference in the average outcome of an NBA game.
Lot to digest but slow down and come back strong with hyping your site. Good Luck

Is there a place to view stats of shot locations of individual games/players? Hoopdata.com is not updated for this season.

I thought people here might be interested in this.

http://www.vantagesports.com/

http://www.kickstarter.com/projects/397 ... the-future

They have really done their research into what information is valuable

1. How did you come up with your dataset?

We track significant actions and outcomes by mixing traditional scouting methods and clearly defined parameters. We worked with the best scouts in the NBA, including Ryan Blake, Senior Director of Scouting Operations for the NBA, and brought in numerous NBA teams and professionals to refine the dataset.

2. Why is it better than SportVu?

It is much more relevant. SportVu is simply a vehicle, just like our technology is a vehicle. The destinations are the data and insights produced. Both vehicles aggregate data - SportVu collects location and movement data only. To even get to our low-level data, SportVu data have to be mined and algorithms created. They are just now able to recognize (most) screens with some accuracy. We already deliver hundreds of new and relevant metrics in every phase of the game.

3. Why is it better than Synergy?

Orders of magnitude more data, far more accurate, better video, more tools.

4. How do you gather all these data?

It's a combination of algorithmic (autonomous) collection and human analysis. Our analysts provide a bulk of the "high-level" analysis (i.e., what constitutes a hedge vs. show in a screen-defense context), whereas we can deduce a lot of the low-level events (scores, game times, possession/chance changes) from computer vision tech. As we continue to advance this technology, we will automate more and more. Our analysts are full-time employees that have all undergone an extensive 3-month training process before they begin submitting data. Furthermore, our NBA analysts have been working for almost 2 years now.

5. How do you ensure accuracy?

Our verification process is a 4-stage system.
1) Our proprietary analysis software prevents analysts from being able to input bad data. It knows, for instance, which teams are on offense and defense, whether the current play resulted in a score, and who scored if so, and therefore, it won't accept input that wouldn't match up with the parameters it understands.

2) Once all players have been analyzed, we have intraplayer checks that verify agreement among all data sources. For instance, if offense_player_1 is seen taking a shot from location "j", then we verify that defense_player_1 (assuming he was defending that shot) also records "j" as the shot location that he defended. If not, the software flags the discrepancy allowing an arbiter to correct it.

3) Finally, we have an inter-rater-reliability system that requests multiple analysis sessions for a certain percentage of plays. It ensures that our analysis is thorough and consistent across analysts.

4) We have built systems to compare our data output with external data sources (i.e., play-by-play feeds) and to alert us to any issues. As you might imagine, we find that we tend to be correct more consistently than official play-by-play.

6. How accurate are your data?

More accurate than play-by-play. A recent audit by the NBA league office showed us to be 99.7% accurate.