Six Nations Rugby

FEATURED AWS SPORTS PARTNERSHIPS

FIVE STATS, SIX NATIONS

First played in 1883 as the Home Nations Championship among England, Ireland, Scotland, and Wales, with the inclusion of France in 1910 and Italy in 2000, the Six Nations Championship is among the oldest surviving rugby traditions and is one of the best attended sporting events in the world. Working with AWS and its partner Stats Perform, Six Nations is adopting new technologies that will help fans better understand the complexities and nuances of decisions made on and off the pitch.

Capture Data

Using cameras on the pitch and broadcast video, AI computer vision and optical tracking senses players, jersey numbers, and the game ball. This data, combined with field zone ID, generates a series of data points that leverage Amazon EC2, S3, and EMR in the capture and storage process.

Analyze It

Edge technology built on AWS analyzes the ingested data which can then uncover formations, playing styles, number of sprints, max speed, distance covered, and more. Proprietary machine learning models running on AWS are continually trained using new data to help recognize individual players.

Distribute Results

Data and insights are generated and served to teams and broadcasters for live commentary and post-match analysis, as well as fan-facing graphic overlays during matches. Machine learning models built on Amazon SageMaker are used to generate the Kick Predictor as detailed below.

Five Fresh Stats

As a contact sport with complex rules and 15 players on each team moving around at once, it can be difficult for even the most seasoned fan to track everything happening on the pitch. The five real-time stats below launched for the 2020 season to help both new and existing fans better understand the split-second decisions made by the players.

Power Game
Visits to the 22
Ruck Locations
Dominant Tackles
Kick Predictor

Power Game
POWER GAME

This stat uses a real-time algorithm to show overall dominance of a team, considering both attack dominance using gainline positive percentage and number of line-breaks, and defense using dominant tackles.

Measuring the direct impact of key event data linked between attack and defense establishes what defines the power of a team.
Visits to the 22
VISIT TO THE 22

Visits to the 22 meter line create more opportunities for a try but also pose a risk. This stat looks at the efficiency of a team’s visits to the 22, including how long they stay in the attack area and how many opportunities are stopped by the defending team.

Measuring the occasions a team entered the opposition’s 22 meter area and how many points they scored with each entry.
Ruck Locations
RUCK LOCATIONS

Using data on ruck location and efficiency, a heat map of the pitch establishes where a team is more or less effective, providing vital insight on an attacking strategy, while unearthing key parts of the field they perceive to be a weakness to the opposition.

Measuring the occasions a team entered the opposition’s 22 meter area and how many points they scored with each entry.
Dominant Tackles
DOMINANT TACKLES

Momentum is key for an attacking team, and defensive pressure can cause a team to lose field position, stopping that gainline momentum in its tracks. This stat measures the percentage of significant tackles highlighting the defensive prowess of a team.

Data is produced through API to be consumable from different resources improving the fan experience.
Kick Predictor
KICK PREDICTOR

Kick Predictor uses historical and real-time data to calculate the likelihood of a kicker successfully scoring on a kick, allowing commentators to not only predict the outcome, but quantify the difficulty of each kick at goal.

Amazon SageMaker is used for real-time inference allowing broadcasters to pull input data needed to come back with a prediction quickly.

PREDICTING KICKS
WITH MACHINE LEARNING

The ability to kick accurately is one of the most critical elements of rugby union, and there are two ways to score through a conversion (worth two points) and a penalty (worth three points). Both are called a “kick” here for simplicity. For a kick to be successful the ball must pass between the uprights.

Kick Predictor is a machine learning model built, trained, and deployed on Amazon SageMaker. The model estimates the probability of a successful kick through the use of historical data such as the kicker’s historical conversion and penalty success rate, combined with real-time, in-game situational data points like game state, kick location timing within the game, and the score at the time of the kick.

The model also outputs the overall historical mean success rate of the players in a given zone. This would be used to compare the real-time success rate prediction of a specific player with the historical success rate for all players.

There are usually 40-60 seconds of stoppage time while the player sets up for the kick, allowing time to flash the Kick Predictor Matchstats on screen to fans, and for commentators to not only predict the outcome, but quantify the difficulty of each kick at goal and compare kickers in similar situations.

This may eventually provide strategic value to teams as well. For example, if a kicker has a lower chance of success from that area of the pitch, the team might look to kick a penalty to the touchline instead – especially if that team is strong at the lineout. Additionally, teams may start to leverage kicking probability models to determine which player should kick given the position of the penalty on the pitch.

Read more +

Less -

TEAM ADVANTAGE

Six Nations Rugby, Stats Perform, and AWS came together to bring the first real-time prediction model to the tournament. The prediction model determines a penalty or conversion kick success probability from anywhere in the field with variables grouped into three main categories.

Location Based

The location-based predictor variables include the X and Y of the goal kick, as well as the distance and the angle of the kick to the cross bars.

Player Performance

The player performance features include the mean success rates of the kicking player in a given field zone, in the Championship, and in the kicker’s entire career.

In-Game

The in-game features include whether the player is playing for the home or away team, the scoring situation before the kick is taken, and in which period of the game the kick is being taken.

Working with the Amazon Machine Learning Solutions Lab

Six Nations Rugby, Stats Perform, and AWS came together to bring the first real-time prediction model to the tournament, which determines a penalty or conversion kick success probability from anywhere in the field. The Stats Perform team leveraged their historical and live data sets and used Amazon SageMaker to build, train, and deploy the base machine learning model with variables grouped into three main categories: location-based features, player performance features, and in-game situational features. The location-based predictor variables include the X and Y of the goal kick, as well as the distance and the angle of the kick to the cross bars. The player performance features include the mean success rates of the kicking player in a given field zone, in the Championship, and in the kicker’s entire career. Finally, the in-game features include whether the player is playing for the home or away team, the scoring situation before the kick is taken, and in which period of the game the kick is being taken.

Model Building

The Amazon ML Solutions Lab worked with Six Nations Rugby and Stats Perform to process millions of historical in-game events data points across 46 rugby union leagues. Data is stored in S3 as JSON format, and is then parsed and pre-processed in Amazon SageMaker notebooks.

Model Training

A wide range of classification algorithms were explored on Amazon SageMaker to predict the probability of a successful penalty kick. The best performing model is a built-in XGBoost model found using SageMaker’s automatic model tuning functionality, also known as hyperparameter tuning. This model has a smaller memory footprint, better logging, and improved optimization compared to the open source implementation.

Model Deployment

The final model was deployed with one-click on SageMaker. During inference, an AWS Lambda function is triggered that queries features from an Amazon DynamoDB table and pre-processes the payload. The predictions are then published through Amazon API Gateway, which is delivered to broadcasters and used to generate on-screen graphics.

IN THE NEWS

Ready to get started?
Learn more about working with AWS Professional Services

Get Started