Would be fun to have something that easily imported into pandas or another decent data platform so one could run some simple regressions/data science on it. Might be able to get some peoplease to help out if you open sourced it on github, or something similar.
If you're doing the collection, and making that data available - I would be very interested.
If it's the framework to do collection myself - not so much.
Have you looked into what the MLB, NFL, NBA, etc would do to block you? While they have no legal protection of facts, you can't copy their facts from them and hope they won't come after you.
No I have not. But is there any way for them to know if a fact was copied from them or not? If Steph Curry made 10 out of 11 field goals in a game. How would they know if I watched the game and found that data out or looked on their website?
I do use APIs for sports data, but I pull that data from existing places. Unless you have some impressive connections or a lot of time on your hands, you're probably not getting data with the same detail as, say, baseball's Sean Lahman or Retrosheet (or commercial options like the data that underlies Synergy video tagging for the NBA), and specialty stuff like SportsVU data (again NBA) is not exactly easy to come by without dropping All Of The Money on a license. I can't speak to the NFL or NHL, but I know datastores exist with accessible APIs and very good data sources.
This is a pretty mined-out space. Might be fun (and if so, go for it!), but I don't know how applicable it would be if you're looking to build it for others to consume.
I don't operate in this space professionally at this point, so I steer clear of the paid options, but everything I listed in that post is a decent starting point. For pure statistics in other sports the Sports-Reference APIs are fine (though I'd rather pull down the data, and the Lahman database is better for baseball); I don't have any good play-by-play data for the NBA aside from commercial sources. I used to have some affordable, but kinda crappy, NBA play-by-play/shot-chart data, but if you're interested in that I'd just be ready to open the wallet.
I don'n really see the point of the question. I doubt you hope to find out there is a significant consumer-base in need, for reasons already mentioned, so are you asking if it's useless? I would say no, because making data accessible always is good.
Personally, I don't care the slightest for stuff like football, hockey or whatever, but I remember struggling to find comprehensive data-set on human physical abilities, such as results in light&heavy athletics, which would include year, age, measurements and all other stuff that might be potentially interesting when analyzing something like that. Data on not-mainstream events is especially hard to find, which is a pity, as it is potentially even more interesting, because olympic athletes are obviously less human-like.
To practice my R skills (I wanted a different dataset to play with) I took an online class from EdX taught by BostonU baseball guru Andy Andres. They talk a lot in the class about sabermetrics and the data sources you can get. The public stuff (either free or through paid sources) is a fraction of the data that the clubs get and use for their own internal use. Things like wind on the field, each pitch speed, where it went in the strike zone, etc. are not released out.
So you have a tough climb to get access to that data without getting the clubs to give it up. But if you build it the fans will come :) there is an amazing number of people that live for the little nuanced data.
For those of you interested in Sabermetrics or think you'd like to know more about baseball stats, I highly recommend Andy's class.
I was looking for an API for NCAA bowl game results for a simple app to coordinate a family pool that we do every year and I couldn't find anything that would cost less than several thousand dollars (I probably wouldn't pay anything really). I just created a dashboard to enter the scores. If something was available I'd love to tie into it.
I currently use sportradar. People will definitely use a sports data API -- but you would have to beat sportradar somewhere (either on developer experience, price, accuracy, feature set, number of sports covered, etc...) to make a viable business out of it.
That's what I was thinking. The easiest way to beat them would be price and developer experience. The price tag of the data is a little pricey and the docs seem hard to understand.
What are the biggest short comings you've found with the service?
You probably won't beat them for B2B. The service is good but not great, but the switching costs are too high that even if your service was better in some way you wouldn't win any customers. They're competent and get the job done and with B2B if you have an existing foothold that's normally enough.
The service is definitely too expensive, the data is occasionally wrong and I have to hound them to fix it, and the developer experience is surely lacking, but at the end of the day a working JSON API of sports data is a working JSON API of sports data, all the other stuff is just noise.
The proper way to beat sportradar would be to find a way to pursue a B2C model instead of a B2B model. i.e. instead of 3k a month to license the data if there was some way to pay $25/month for a similar experience you might be able to capture a much larger market share that sportradar prices out. If you could get college kids hacking on sports stats to use your service because it's the only thing they can afford you might be able to grow that into something.
Yeah thats what I was thinking, a very low entry point. I don't see any problems associated with only charging $15-25/month to access the data.
One thing that I have been wondering is, how do I get away with using the data without the leagues intervening? Or do they not own the data because the stats are publicly accessible
Absolutely. I have a couple of sports apps [0] that use both scraped web data and a hidden API to get realtime baseball game information. One of them won a sports hackathon a couple years ago (sports radar was a sponsor).
From what I recall, sports radar is really expensive because they have people manually entering data all the time. If you could find a happy medium where your data is close to real-time but doesn't require a lot of manual work, you could keep your costs low and prices affordable for small app developers.
I'd suggest targeting mobile developers and not just fantasy/daily fantasy. I'd love to work some more up-to-date questions in my sports trivia app. Example: "What had the most rebounds in Warriors/Cavs game 1?"
Absolutely. I have a couple of sports apps [0] that use both scraped web data and a hidden API to get realtime baseball game information. One of them won a sports hackathon a couple years ago (sports radar was a sponsor).
From what I recall, sports radar is really expensive because they have people manually entering data all the time. If you could find a happy medium where your data is close to real-time but doesn't require a lot of manual work, you could keep your costs low and prices affordable for small app developers.
I'd suggest targeting mobile developers and not just fantasy/daily fantasy. I'd love to work some more up-to-date questions in my sports trivia app. Example: "What had the most rebounds in Warriors/Cavs game 1?"
That's what I was thinking. I'm sure live games you have to manually enter but other than that it can be automated. And I really want to help mobile developers and developers in general. There are some great apps that can be built out there. They just need data access.
You're probably better off with an OCR system that's configured per channel display format, rather than an ML solution. ML systems require training regardless of effectiveness.
Yes indeed. I have an ongoing project with college football and I was surprised nothing good for this actually exists.
Correctness and comprehensiveness is probably more important in my mind than having the information immediately. I'd wager most people around here would want to use it for training data rather than any sort of real-time system.
I was just trying to start with really simple things like all of the weekly game scores and what rank the teams held at that point in time. I was surprised no one had this and I just had to write a scraper.
Definitely! Thinking about building a natural language query API over sports data, similar to Statmuse [0], but the whole thing kind of depends on having data in the first place.
Individual player and team stats, probably at a per-game resolution, would be ideal. Relational works great but any format could work. Additionally, a join table of player-team memberships over time (per season?) would be useful for visualizing timelines. Bulk data downloads could also be useful, since the tool I'm envisioning wouldn't depend so much on real-time data as the ability to explore past trends.
Let me know if you need any help. We built a machine learning service called hopdata and I have been looking for a sports API to link into so we could do something similar to fivethirtyeight.
http://www.hopdata.com
At this point, I had it in my mind as a broad and large undertaking, we would really need to dig into what information is available. For instance, if we look at baseball, we could make predictions based on historic outcomes and real-time scores but it would be much better the more granular we can get. Current score, current inning, current outs, current count, which players are in the outfield, who are the next 9 people to bat, who is pitching, which ballpark, the weather...
Once in a blue moon I get the urge to run some analysis or another on sports stats, but I almost inevitably end up having second thoughts when I realize obtaining good data is going to be an ongoing chore.
On the upside, the existence of that friction definitely points to a gap where value can be added. Another likely source of friction where you can add value is the work anyone working on multiple information sources (whether this is several databases/apis for one sport, or for different sports altogether) has to do to integrate multiple services and data models. If you can identify some people who are doing that work in the first place, you can probably make a business case for how you can simplify their application and give them a single interface/API for all of that data.
I'll also echo the suggestion from @scottrblock on the potential usefulness of various metadata. If all you have are game statistics, while people can obviously do lots of novel analysis on the data, it's fairly obvious what the "use" is. With a wide spectrum of metadata, you might end up simultaneously creating something of value to someone analyzing the effect of weather on different sports/teams/players and for hotel chains trying to set prices based on the historical attendance for all sporting events within 50 miles.
If you could create a revenue model that supports delivering the service both to continual high-volume business users and to users just using the API occasionally to support an analysis project, perhaps it's possible to pass revenue downstream to other data sources. Granted, this adds a number of administrative problems (people submitting data collected/generated/sold by other parties as their own to collect revenue on it; getting low-quality/falsified data; etc.)
* I wasn't aware of stattleship or sportsradar before this thread; no idea how well they cover these.
53 comments
[ 4.9 ms ] story [ 113 ms ] threadAs you can tell, most of my work is in python. But R is absolutely huge, so providing an interface to that would certainly be handy to most.
If you're doing the collection, and making that data available - I would be very interested. If it's the framework to do collection myself - not so much.
Looking forward to more details!
This is a pretty mined-out space. Might be fun (and if so, go for it!), but I don't know how applicable it would be if you're looking to build it for others to consume.
Personally, I don't care the slightest for stuff like football, hockey or whatever, but I remember struggling to find comprehensive data-set on human physical abilities, such as results in light&heavy athletics, which would include year, age, measurements and all other stuff that might be potentially interesting when analyzing something like that. Data on not-mainstream events is especially hard to find, which is a pity, as it is potentially even more interesting, because olympic athletes are obviously less human-like.
So you have a tough climb to get access to that data without getting the clubs to give it up. But if you build it the fans will come :) there is an amazing number of people that live for the little nuanced data.
For those of you interested in Sabermetrics or think you'd like to know more about baseball stats, I highly recommend Andy's class.
What are the biggest short comings you've found with the service?
The service is definitely too expensive, the data is occasionally wrong and I have to hound them to fix it, and the developer experience is surely lacking, but at the end of the day a working JSON API of sports data is a working JSON API of sports data, all the other stuff is just noise.
The proper way to beat sportradar would be to find a way to pursue a B2C model instead of a B2B model. i.e. instead of 3k a month to license the data if there was some way to pay $25/month for a similar experience you might be able to capture a much larger market share that sportradar prices out. If you could get college kids hacking on sports stats to use your service because it's the only thing they can afford you might be able to grow that into something.
One thing that I have been wondering is, how do I get away with using the data without the leagues intervening? Or do they not own the data because the stats are publicly accessible
From what I recall, sports radar is really expensive because they have people manually entering data all the time. If you could find a happy medium where your data is close to real-time but doesn't require a lot of manual work, you could keep your costs low and prices affordable for small app developers.
I'd suggest targeting mobile developers and not just fantasy/daily fantasy. I'd love to work some more up-to-date questions in my sports trivia app. Example: "What had the most rebounds in Warriors/Cavs game 1?"
[0]
https://itunes.apple.com/us/app/hat-trick-daily-sports-trivi...
https://itunes.apple.com/us/app/on-deck-baseball-alerts/id91...
From what I recall, sports radar is really expensive because they have people manually entering data all the time. If you could find a happy medium where your data is close to real-time but doesn't require a lot of manual work, you could keep your costs low and prices affordable for small app developers.
I'd suggest targeting mobile developers and not just fantasy/daily fantasy. I'd love to work some more up-to-date questions in my sports trivia app. Example: "What had the most rebounds in Warriors/Cavs game 1?"
[0]
https://itunes.apple.com/us/app/hat-trick-daily-sports-trivi...
https://itunes.apple.com/us/app/on-deck-baseball-alerts/id91...
Correctness and comprehensiveness is probably more important in my mind than having the information immediately. I'd wager most people around here would want to use it for training data rather than any sort of real-time system.
[0]: https://www.statmuse.com/
I'll also echo the suggestion from @scottrblock on the potential usefulness of various metadata. If all you have are game statistics, while people can obviously do lots of novel analysis on the data, it's fairly obvious what the "use" is. With a wide spectrum of metadata, you might end up simultaneously creating something of value to someone analyzing the effect of weather on different sports/teams/players and for hotel chains trying to set prices based on the historical attendance for all sporting events within 50 miles.
If you could create a revenue model that supports delivering the service both to continual high-volume business users and to users just using the API occasionally to support an analysis project, perhaps it's possible to pass revenue downstream to other data sources. Granted, this adds a number of administrative problems (people submitting data collected/generated/sold by other parties as their own to collect revenue on it; getting low-quality/falsified data; etc.)
* I wasn't aware of stattleship or sportsradar before this thread; no idea how well they cover these.
I've been looking for something like this for one of my personal projects, but, haven't really had the time to spend on it recently.
I imagine William Hill etc would find it useful