Deep Learning is Coming to Hockey

deep learning hockey
MONTREAL, QC - NOVEMBER 05: The NHL logo on the back of the goal netting between the Montreal Canadiens and the Boston Bruins at the Bell Centre on November 5, 2019 in Montreal, Canada. The Montreal Canadiens defeated the Boston Bruins 5-4. (Photo by Minas Panagiotakis/Getty Images)

Analytics have been transforming how we watch hockey. The revolution is just beginning. Statisticians and quantitative experts have led the way. Their impact has changed how we discuss and watch hockey. Analytics have been influential. Deep learning will be disruptive.

Advances in computing and understanding of complex relationships will massively alter the sporting landscape. Hockey will not be immune.

Every decision point is potentially affected. This will lead to impacts on and off the ice. Whoever gets there first will have an enormous competitive advantage. Think Moneyball, but with a team that maybe doesn’t lose in the playoffs. 

Deep Learning in Hockey: It’s Coming to the NHL

Computers Learning

Our technology is getting smarter. Deep Learning (also known as machine learning)  is coming to many aspects of life. The basic idea is using a computer to analyze complex interactions to come to conclusions. We have seen the concept applied to medicine with great results. The world’s greatest GO player has left the game after realizing the robots can’t be beat. Team sports will be conquered next.

How it Works

High-end computers can do mathematical calculations we humans can only dream of. This is the basis of how it can work.

Machine learning is an application of Artificial Intelligence (AI.) The focus is providing data to computers, which then learn and improve with experience. These machines aren’t programmed in the traditional sense, rather they are developed by allowing computers to access data and learn from it themselves. 

The Forefront

Like in the outside world, the impacts for sports are numerous. There are many potential applications for deep learning. A look at the call for papers for the 2020 Machine Learning and Data Mining for Sports Analytics conference shows what this world is working on. Expected topics include items such as:

  • Tracking data (positional information)
  • Strategy/Tactics
  • Player valuation and acquisition
  • Training regiments
  • Injury prediction and prevention
  • Outcome prediction.

A quick glance at the topics demonstrates the field is getting into increasingly complex issues. This has the potential to reshape coaching, management, and player development. 

The Problem for Deep Learning in Hockey

There is good data and bad data. Like the larger debate about analytics, the availability and value of information is of concern. The sheer number of variables in the chaotic environment on the ice makes the analysis complex. Stop and go sports like baseball and football are easier to analyze as the statistics tend to be more clear cut. 

All numbers aren’t created equal. The issue of inconsistent stat keepers will slow progress down. A shot or a hit in one arena may not be the same in the next. Stats also become less reliable away from professional leagues, and so a close look at the numbers going in are needed to produce accuracy. Quantitative analysis is wonderful, but critical analysis to ensure accuracy is needed. In science speak, you need to operationalize things properly. 

The complexity of hockey will make adopting deep learning difficult. It will be one of the last sports to truly be able to take advantage of it. There are many ways it will affect the game for fans, players, and teams. The complexity problem will be overcome. 

Prediction Will Come First

Who’s going to win? Can statistics help us understand the answer? Apparently, yes.

Predicting results has been a primary focus of deep learning applied to sports. The first tests have focused on predicting results. The potential of figuring out who’s going to win, and how to efficiently bet would be lucrative for outsiders. Like in other sports, this is the first area where deep learning is likely to come. 

It has been a long road, but expert pundits are falling. In the early days of deep learning, the “experts” at prediction on tv were better. This is changing. Back in 2003, early attempts computers were not able to beat expert pundits at prediction. Recently, a deep learning machine (75% accuracy) was able to beat the ESPN team’s 63% accuracy over the same time. This is just the first step. 

The Spread of Deep Learning in Sports

Football experts were the first to fall. Machine learning will change the game well beyond that. They have the ability to be early adopters in the field. Particularly as the NFL has so much money, they are likely to continue to be the league to watch for the effects of deep learning. 

That said, this is spreading. It has been applied to the English Premier League and many other sports. When it arrives in the hockey world, it will change how teams manage their decision making at all levels. From who to sign as a free agent, to who to trade for, and even lineup decisions night to night. The applications are limited only to the availability of the data.

Chayka You Money Makers

While hockey is chaotic and numbers are inconsistent, this problem can be lessened. Stathletes seem likely to be the people who do it. Hockey is well aware of the name Chayhka already. Meghan is the one to watch in this case. She was one of 3 co-founders of the company along with brother John and Neil Lane.

What they do:

Using proprietary video tracking software, Stathletes pulls together thousands of performance metrics per game and compiles analytics related to each player and team. These analytics can provide baseline benchmarking, player comparisons, line matching, and player and team performance trends. Stathletes currently tracks data in 22 leagues worldwide and sells data to a wide variety of clients, including the National Hockey League (NHL).Via FedDev

If they are using machine learning, it is not clear. If not, it seems inevitable that they will. Meghan Chayka currently works with an expert in machine learning at the TD Management Data and Analytics Lab at Rotman (business school) at University of Toronto. Seems likely they can benefit each other, and would know this. (This may be part of the reason why Arizona seems peeved at Chayka currently. They may have just become a data have not.)

Stathletes and other groups are gaining knowledge and information. They will improve as they go. The NHL is open to this, it’s coming.

The Short Version

Machine learning has arrived. As the ability to obtain information improves, it will coincide with further developments and what’s to come. If you are able to follow, Neil Lane (current Stathletes CEO) is to speak at the University of Waterloo on what sports managers can learn from analytics. This should be enlightening. 

Embedded items will be key. Chips and sensors in various hockey items are coming. Jerseys and pucks will be transmitting the information. Learning computers will put it together.

The impacts will be numerous. Coaches, players, agents, and teams will have considerably more knowledge. This changes decision making. Training. Diet. Trades. Penalty Kill lineups. The possibilities are endless.  

Deep learning will lead to hockey having more knowledge of all aspects. If people like Pierre McGuire hate analytics now, just wait for what’s to come.

Main Photo:

Embed from Getty Images

1 COMMENT

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.