I'm Mike Hatfield, and welcome to my tracking data visualizationwebsite. I'm a data scientists with experience in computer vision and machine learning, and an entusiasm to waste my time watching any spors I can. Making this website has been a semi-productive way to combine the two. Read through any of the below sections to learn more about how this website has been developed and structured, the data processing, the animation, and a handful of other research questions I've tried to explore.

This website is built using a combination of AWS resources. An EC2 instance to support the flask server, an S3 bucket to store the data for display in the lists and animation, and Route 53 for handling the domain name routing and health checks.

Earlier iterations of this website started with less scalable and resilient architectures. Data storage originally began by simply storing the json files foreach play in the EC2's local file directory from where the flask server is run.

You can see in the Future Development section that there's additional work planned to improve the cloud architecture. The next focus in this area will be creating health checks for the EC2 instance (rerouting to a static S3 webpage when unhealth). And if this website ever gets visited at a reasonably regular volume, I'll need to build out a structure using Load Balancers and Auto Scaling Groups.

The data visualized in this website pulls from two sources. The first is directly from the data supplied by the Kaggle Big Data Bowl 2021, which only includes player tracking data for those involved in passing plays (i.e.: no OL or DL unless playing in coverage), and data starts with the hike, and ends with the dead ball call. The second is player tracking data pulled directly from sportradar for 5 test games. The player tracking data for these games includes all 22 players, and data starts at the huddle, and goes well past the dead ball call.

With those differences in mind, the tracings for the Big Data Bowl games looks quite clean. However, the tracings for the “raw” datasets, look a bit messier, seen here (Q1 (4:59) (Shotgun) J.Goff pass short right to T.Gurley for 19 yards, TOUCHDOWN):

These 5 test games are from 2018 so it’s possible the metadata for each play has been updated since, however, they don’t include timestamps for the hike or dead ball. But we can successfully automate that process using existing change point detection algorithms. With a python library called ruptures, we can plot the ball’s velocity (y-axis) against time (x-axis) to get a view that looks like this:

The blue and red sections represent the labeled points for the hike (end of blue) and dead ball (end of red) times. The dotted lines are the predictions from the change point detection algorithm. You can see in this representative example that the hike time is accurately predicted, and the dead ball time is closely predicted, which will be good enough for visualizing the animation tracings for our purposes.

Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.