Students create model for projections and optimized lineups for daily fantasy sports
The first fantasy sports league was created 38 years ago and grew rapidly from there. Today, according to the Fantasy Sports Trade Association, fantasy sports is a $7.22 billion industry. Around 11 years ago the first websites to specialize in daily fantasy sports appeared. Now they are worth around $1 billion dollars.
There are several differences between fantasy sports and its subset known as daily fantasy sports. Fantasy sports is where a league of players, usually a group of friends and/or coworkers, choose players to assemble a fantasy team. This version is based on the entire season of the sport and points are won when the season is over. Daily fantasy sports is played online with thousands of other players. Instead of waiting an entire sports season results are released each day. Daily fantasy sports also differs in that it focuses more on a monetary prize associated with winning. This prize can reach as high as one million dollars per contest.
Using 42’s project incubator, a group of students are creating a website subscription service that they have named Project Alea. The name will change to something more specific in the near future. Alea was chosen because it is Latin for a game of chance. Jarvis Nederlof, the project manager, has a background in sports and finance. He was inspired to combine his two passions into trackable data. With Erik Rintala and Sam Collet on the team, they are able to make player predictions using machine learning and a neural network. First applying their model to hockey they are now using it to make predictions in baseball. The team is hoping to expand their service to several different sports, including the esports scene, in the near future.
What is the purpose of your project?
Jarvis: The purpose of the project focuses on daily fantasy sports. Daily fantasy sports has been growing a lot in the past few years. We are building a tool which utilizes the current technology of neural networks and complex solvers and our goal has two parts. Our primary focus is to remove any bias associated with picking players using our projection system. We then pair our projections with our customizable optimizer which generates the best possible lineups for a user.
How does the model you created for daily fantasy sports lineups work?
Jarvis: We have a bunch of different data service providers that we pull our data from. We have all of this data that we pull down and we organize it in a way that we are able to make a prediction for every single player that will play that day. Right now we are just doing baseball, but we have a previous model for hockey.
We use a neural network to make our projections because it is so complex and difficult to predict how a player will do. A neural network is great at finding the stats that matter. It is so easy for someone to just pick their favorite player or someone they see in the news. The goal with our projections is to remove a user’s bias and give them a more accurate pick.
The second part is we take all of the projections that we have and we run it through an optimizer to build our lineups. The idea with the optimizer is to generate up to 150 lineups of increasing variance. The ultimate goal is to provide the user with a better contest experience where they walk away with a bigger share of the prize. The optimizer works by allocating players onto lineups based on a bunch of different constraints. The most basic constraints are the salary cap and positional constraints. We then add a bunch of our own constraints to make the model better. By putting all of these constraints together we help users build better lineups. We also solve the problem of putting players together to maximize points.
Sam: We will be adding others. E-sports is something we are interested in adding. The concept is the same because esports and physical sports are both in a competitive environment. But esports contains a different skill set because it is based on people playing video games against each other and the actual results of the game.
Jarvis: Esports is growing a lot and it will be an extension of what we are doing.
Describe the work you did on the project:
Sam: I guess I would say I did continuous integration. I touched on every aspect of the project in some way. My biggest contribution would be setting up the structure and focusing on the ecosystem and scalability. Right now I am working on the website. I am interested in getting this project to scale, especially when we want to grow a customer base. Jarvis added, “Sam has been more of a forward thinker when it comes to the tech side of it.”
Jarvis: I was the one that started the project. Before we came together I built this kind of system for hockey. The idea was to bring in more people to scale it and bring in more sports. I have the domain knowledge, sports are my background and interest. Because of that, I am able to make a lot of decisions when it comes to getting the data. I worked with Sam to build the neural network and to get the optimizer working. Those are the cornerstones of what we offer so I have been focusing on those two elements of the project.
Erik: I’ve been more focused on the infrastructure so the tech that kind of underpins the program process of what we are building. So I have been working a lot on the database and servers and testing and deploying new features.
What was the most difficult part?
Jarvis: I think from my perspective was managing the team. I had to learn a lot that I didn’t know and that I didn’t expect that I had to learn. The tech side is where we have encountered problems but we are able to solve them. I had to learn a lot about project management.
Sam: The hardest part for me wasn’t the technology, you have the bumps but you know where you are going. I worked on teams before but there are always new dynamics that you have to figure out.
Erik: Time management. When I have finals (I am taking online classes to finish a CS degree) it is really difficult. I had to get my projects done in school and make sure I was not holding back our project.
What did you enjoy most about your project?
Jarvis: It’s been the machine learning and neural network of it in greater detail, and we are getting really good at what we are doing. The industry and deep learning are really growing and booming right now and it is exciting to be a part of that.
Sam: I would say probably learning a little bit of everything. For me it was definitely data engineering, managing the data in general, I could say that about any aspect of that project. Looking into the future and seeing what we can do more with the project is exciting.
Erik: I liked the feeling of when we encountered an obstacle and worked through it, and that feeling of success when you finally figure something out.
What did you learn from this project?
Jarvis: Team management, working with a team in tech, and the new technology like database management and neural networks.
Erik: It’s a lot around teamwork. I think I have a tendency to think super far ahead and get caught in tiny intricacies, while Jarvis balances it with a pragmatic approach and seeing what happens.
Sam: We had to create a team that can generate a product, and that in itself is a learning experience.
What future do you see for your project?
Jarvis: We are close to discovering what the future will be, it is hard to know what that exactly is right now. We are developing a fantastic product. Before we finish there is a lot that needs to happen. So the challenge will be getting it to a point where we can share it with others. Sam added, “We want to build a user base.”
Meet the Team
Name: Jarvis Nederlof, Project Manager
Hometown: Red Deer, Alberta, Canada
Interests: I was always interested in sports, I played hockey for 20 years competitively. Other than sports I really like data and manipulation of data, I am pairing these interests together now.
Dream job or long-term career goals: I would love to run my own company with my brother and work with him to build up something that I can be proud of and share with other people. My family has a history of running a business and I would like to build on that.
Name: Erik Rintala, Lead Developer
Hometown: Issaquah, Washington
Interests: Coffee, video games, machine learning. I like understanding the foundational parts of technology, we are working with machine learning and AI and I have a hard time jumping in without fully understanding. I am a fan of technology and reading about tech.
Dream job or long-term career goals: I think it is probably going to be in machine learning and AI.
Name: Sam Collet, Software Developer
Hometown: Albany, Oregon
Interests: Video games got me into programming in high school, so I am interested in video game design and how people approach interactive environments.
Dream job or long-term career goals: Definitely independent contract work, hopefully with an independent contracting company.
Project Alea Graphic Designed by Melissa Egan
published by Stacey Faucett – June 14, 2018