Generic filters

Is this video software company set to take hockey analytics to the next level?

5 years ago

What actually happened? What was actually important?

Those are water cooler questions that go back a long time. They’re also questions that hockey’s number-crunchers have been chasing for some time now.

The challenge in answering those questions, since day one, has been the quality of data.

But now a team of computer scientists, led by a figure skater from North Vancouver and a computer scientist from Iran, may finally have figured it out?

A host of hockey people and famous sports owner and tech billionaire Mark Cuban think so.

Craig Buntin knows so.

Buntin is the CEO of SportLogiq, the Montreal-based tech startup which is looking to change the way coaches, scouts, players, fans and broadcasters look at the game.

“Reaction’s been everything from ‘My God this has been what I’ve been waiting for for ten years’…the other half is ‘this is a lot of data and we don’t know what to do with it,'” he said. But, there’s no doubt, Buntin believes, the enthusiasm is there and his team is ready for the challenge.

Other sports are using high-tech cameras or radio-frequency ID chips to track player movement and record game data, but SportLogiq has its own twist: computer vision and machine learning.

Buntin, who grew up in North Van before finding his training home in Kelowna, founded SportLogiq with Mehrsan Javan, who now serves as the company’s Chief Technology Officer.

Javan completed his PhD at McGill’s Centre for Intelligent Machines, where his research focused on statistical modelling, computer vision and machine learning. The technology SportLogiq now employs grew out of Javan’s PhD thesis.

“The idea starts and then you meet smart people,” Buntin told The Province over the phone this week from Montreal.

SportLogiq has developed software that can process submitted video, look for patterns it has learned from previously-viewed data and then code it for further analysis. How to use the data was a question from the beginning.

“Very early on we brought on a guy named Chris Boucher, who spent years in his house manually going through games, event by event. He came up with a very labour-intensive method,” Buntin explained. “We had a template to start with. Then we brought in the computer vision tech and the engineers and we said ‘we can automate this.’ ”

Automated mass data collection is SportLogiq’s secret sauce.

SPORTLOGiQ Technology from SPORTLOGiQ on Vimeo.

Buntin spent a decade at the top levels of figure skating. He finished 11th in pairs at the 2006 Olympics in Turin, skating with partner Valérie Marcoux. The pair were national champions three times as well.

After retiring from competitive skating in 2010, Buntin studied for his MBA at McGill. His interest in hockey also drew him into the emerging world of sports analytics.

Hockey analytics was growing by leaps and bounds but a problem persisted: There were always gaps in the data. You could say “this guy is good at this” and you could look at other numbers to figure out much of that player’s skill set, but it was still an incomplete picture.

If a player gets a good scoring chance, is he solely responsible for that chance — or did a teammate’s win of a puck battle contribute as well? Are some players better at retrieving loose pucks over others? Do we actually know who the best passers are?

There’s also a significance problem — not every data point is equally useful. When you collect large volumes of data, a lot of it doesn’t tell you anything at all.

That’s been the biggest challenge to the analytics revolution in hockey — determining which events and skills really matter. Players can be have poor puck-possession numbers, but how much is that about them and how much is that about their deployment on the ice or the quality of their teammates.

Since the SportLogiq system captures everything, Buntin believes they’ve solved that problem. They can build a picture of what’s happening at any given moment in a game and then break it all down to see what happened.

“We are not here to make a small impact, we’re here to fundamentally change it,” he said. “Once you can locate every single player, every single joint on the ice, everything can be quantified.”

Note he said “joint.” Yes, this system can track not just player position, but the player’s individual body movements. The most effective techniques can be determined. It’s a coach or scout’s dream.

The premise is simple: Video, it doesn’t matter where from, is fed into the SportLogiq software, which has been programmed to recognize human movement and record all the events presented to it. Machine learning allows the computers which are processing the video to recognize from past experience what it’s seeing in the new, never-before seen video.

The computers code the video and records events. An average game has between 2,500 and 3,000 events, Buntin said his team has found. Once the software has processed the video, the data are double-checked by the SportLogiq team to make sure nothing’s been missed. With all this new data, they’ve looked at stats and skills which conventional analytics has already revealed to be useful and are now finding new ones as well.

One of the skills SportLogiq has uncovered is “loose puck recoveries.”

“It’s a combination of speed, strength and hockey sense and has a huge impact on scoring,” said Andrew Berkshire. Berkshire was the lead writer and editor for noted Canadiens blog Habs Eye on the Prize, but was hired this summer to manage editorial content for SportLogiq’s site. Data-focused hockey writer Thomas Drance has also been brought into the fold as a freelance contributor.

“Loose puck recovery has a huge impact on scoring, we’ve found,” Berkshire said. “It’s a data point that emerged because the whole picture was now available

“We’re looking at what’s the most common outcome when a player does ‘x,’ ” he said. “Another was looking at whether certain players were more effective in throwing hits that turned over possession.

“Alexei Emelin had many hits that took guys off the puck but few that turned over possession,” he said. A famous teammate of his was far better: 65 per cent of PK Subban’s hits led to a change in possession.

The reception in the industry, from teams and analysts alike, has been positive.

“We’ve been working with about a half-dozen NHL teams in the last five or six months. We provided data on at least half of the players who were drafted,” Buntin said. Two broadcasters have also expressed interest.

And then there’s Mark Cuban, owner of the NBA’s Dallas Mavericks who made his billions in the tech industry and has long held interest in the development of sports analytics.

Buntin and Javan started their funding search a year and a half ago with Montreal-based startup incubator TandemLaunch, who told them to go out and find the people, on the tech side, on the communications side and on the development side.

“When we first put the fundraising plan together, we set a target and went after smart money. Cuban was top of the list. Having him on board, I knew would be big from a product and market development standpoint, but also for networking in general,” he said.

As soon as Cuban saw what Buntin’s team was working with, he was impressed. That SportLogiq’s technology will be available to anyone with a camera — not just professional teams — is what sold him, Buntin said.

“It’s the thing we offer that no other tech does,” he said. “Other systems use expensive multi-camera setups or wearable chips.

“We said, if you could do what they do but with just one camera, you could change all this. You look at every single sport, at every level, there’s at least one camera. We can bring what the pros have to everybody.”

Last month, Cuban put $1.7 million in seed money into the company, alongside a number of other new investors, so SportLogiq could bring more talent on board.

The company is hiring computer engineers left, right and centre because interest is so great.

“What we’re doing first is really exploratory,” Buntin said. “We’re coming up with every thing that’s possible to look at in a game, and then looking at how the data clusters. There are really clear initial stories that pop out.”

Note: This article originally appeared in The Province. Click for link here.



Leave a Reply