Fantasy football and artificial intelligence
It’s not a good idea to speed-read everything online about your fantasy football teams and players. Maintaining focus among the deluge of information will, at best, result in human data amnesia and, at worst, result in the total collapse of your fantasy football squad. It’s impossible for humans to comprehend the limitless amount of knowledge, let alone make inferences from different sources. Decisions that lack sufficient data knowledge are rife with uncertainty. You may concentrate on making important decisions, such as who to start and which players to trade, with the aid of artificial intelligence (AI) tools.
The straightforward and complementary user interfaces of Player Insights with Watson and Trade Assistant with Watson can help you handle the data and decision space tsunami and increase your confidence in your choices. The player potential is outlined in Player Insights with Watson. The comparative floors and ceilings amongst players are easily ascertained. In parallel, Trade Assistant with Watson identifies transactions that will improve your team’s overall performance. The easy mobility of players within a league becomes your new advantage once these two systems are in place.
Both Player Insights with Watson and Trade Assistant with Watson are powered by AI running on a hybrid cloud, enhancing your chances of winning while letting you enjoy fantasy football. Let’s investigate how.
AI in entertainment and sports
The computing industry has produced mature technologies that support AI over the past few years. Some well-known instances that have influenced AI algorithms are debating, transaction optimization, and prescriptive methods. Over the past several decades, IBM® has taken the lead in the partnership between business, academia, and the government to develop machines that can supplement human skill using a variety of AI algorithms. Figure 1’s pyramid of AI capabilities represents the fundamental components of cutting-edge systems. The technology powering IBM Project Debater, cloud-based IBM Watson® Services, and IBM AI research initiatives enables the ascent up the pyramid. These ground-breaking initiatives each developed innovative technology for optimization with neural approaches, a brand-new and developing sector.
AI pillar technologies are included in the pyramid’s initial phase. Here, we concentrate on descriptive statistics, including histograms, means, modes, and variance. Many of these count statistics are employed in correlation, significance, and hypothesis testing procedures. We launched SlamTracker at Wimbledon, a real-time score application with extensive analytics and data. Players are grouped according to performance levels and player types for the Keys to the Match feature. Then, we retrieve information about the statistics that are now being scheduled. Applying decision trees to figure out what strategy a player should use to maximize their odds of winning allows us to start moving up the AI capabilities tree.
Predictive modeling is the next phase, and its methods can be broadly categorized as a form of machine learning. Traditional support vector machines, clustering based on density, and forecasting models are all examples of predictive modeling. We used predictive models to assess the total excitement level of video highlights from the Masters over the past three years. A support vector machine and an xgb-boost algorithm ensemble, in particular, determine the overall level of enthusiasm. Using video streams, we build candidate highlights and rank the excitement using game context, gesture, and sound predictors. All of this has to do with predictive modeling.
Natural language processing technical breakthroughs represent the third level of the pyramid. IBM has been in charge of this, and you may be familiar with their Project Debater and Jeopardy! Challenge. Through application, AI has rapidly matured in this field. We employ a number of fundamental NLP techniques, including text summarization, evidence detection, and explicit semantic analysis. We used Match Insights with Watson and Open Questions with Watson at the US Open 2020 to interact with the public. We asked the crowd a several of questions, including, “Did Pete Sampras have the best serve-and-volley game?” Fans would offer their thoughts on the matter. The quality of the argument and whether it is a pro or con argument are then determined using natural language processing. The feedback was then incorporated into the narration to reinforce the initial assumption regarding Pete Sampras.
The resurgence of neural modeling is the next step in the AI capacity pyramid. We have made numerous contributions to the field. For the 2020 GRAMMY Awards, we used informative factoids that have to be brief stand-alone facts about the artist to contextualize their nominations. We tested if a statement was insightful using models like the Bidirectional Encoder Representations from Transformers.
Prescriptive optimization now occupies the top position on the pyramid. A large portion of the pyramid’s strategies serve as foundational elements for prescriptive outcomes. Trade Assistant with Watson in ESPN Fantasy Football offers consumers a trade strategy. While reducing player cost, we maximize player value. It was possible to use this consumer-facing application for the whole 2020 Fantasy Football season.
The results of our projects in the sports and entertainment industries serve as examples of how AI is used in practical settings. However, let’s now concentrate on fantasy football.
Fantasy football using AI
Over the past four years, we have collaborated with ESPN to support fantasy football managers at all skill levels in making informed choices. Through this project, we paved the way for how customers use AI.
At the IBM labs in Austin, Texas, in the early half of 2017, we had a meeting with ESPN. We started brainstorming using the art of possibility, or as they used to say in the labs, “throwing mud on the wall and seeing what sticks.” A few IBM engineers went out to dinner on the first night and drew up an architecture centered on natural language processing tools. To make sure I wasn’t crazy, I showed the back of the napkin sketch to my other IBM coworkers. Fortunately, we are still sane today (at least I think). Then, we presented our ideas to ESPN. We collectively made the decision to create a system that could read the “internet” during NFL seasons in order to forecast player statuses and performance trends. We had no idea if it would work or be precise enough to assist managers in making important decisions. It turns out that our technology consistently surpassed accuracy standards throughout the process.
We took on the challenge of a covert release onto the ESPN Fantasy Football desktop experience in the second part of 2017. We used an iframe tab to display a very straightforward player card. Consumers saw our player insights for the first time right now. We displayed boom and bust percentages, playability probabilities, and the likelihood of having a concealed injury. Even score distributions were displayed. The overall system was approved for the 2018 season following user testing and feedback.
Once everyone had signed on for the 2018 campaign, we added our app to the ESPN mobile interface. Our pipeline was now able to ingest movies and podcasts to extract insights from in addition to the user experience. Now Watson may view videos and listen to podcasts to learn more about player performance. According to some of the feedback we received on our user interface, people appreciated the feature that allowed them to compare players’ results visually. Our work had an unparalleled rate of acceptance as a result of these feature improvements. Traffic to our experience was rapidly increasing week over week.
2019 brought more traffic than we could have ever dreamed. The majority of our daily offering of over 2 billion AI insights came from North America. Debiasing algorithms were included for this iteration to ensure that player booms and busts were distributed equally among teams, regardless of how well-liked the teams were. With all these state-of-the-art improvements being made to our unique system, a query started to surface that could only be resolved through advancements in AI technology. How could we design an AI-based experience specifically for you?
We developed Trade Assistant with Watson because, as the phrase goes, a content league is an active league. During the entire 2020 season, the function was accessible. Your proposed trades are determined by the roster, regulations, and competition of your league. Due to how highly subjective trade evaluations are, this project is also incredibly difficult.
Player Insights, Trade Assistant, and Watson all work together as part of the system’s integrated AI design.
Let’s examine the Player Insights system first. The Player Insights system continuously processes players through the AI pipeline in batch and crawler modes to provide you a greater insight of your players. Through various machine learning procedures, the Python code generates output and writes to an IBM Db2® on Cloud database. Insights are generated by a Node.js content generator and uploaded to our origin, IBM Cloud Object Storage. Customers can access the data from this point via the IBM Content Delivery Network. Player Insights and Trade Assistant’s integration points are Db2 and the Content Delivery Network.
To build the broad variables we previously indicated, the Trade Assistant batch task gathers information from Db2 and the Content Delivery Network. The 525 Red Hat OpenShift Kubernetes Service pods that are accessible on demand can access this data because it has been uploaded to IBM Cloud Object Storage. Every five minutes, each pod updates its memory with the most recent price and valuation information. We generate trade packages and send them back to ESPN for screening when requests come in from fantasy team owners.
All user traffic to our scaled-out OpenShift clusters or to our batch jobs’ AI insights is routed through our Content Delivery Network. Customers can so seamlessly switch between the ESPN product and the two IBM projects.
Large-scale AI
Watson’s Player Insights and Trade Assistant are complementary systems. They both utilize the Red Hat Open Shift Kubernetes Service and are AI pioneers. However, there are two distinct variations in the computational request paradigm between the two systems.
Every night, Player Insights with Watson processes batch workloads to generate AI data about the top fantasy football players. Our crawler software is then active to check whether there have been any changes to the player’s status since the batch run once the batch job has finished for all of the top fantasy players. The player’s expected score may have changed, they may have been placed on/off injured reserve, suspended, or even given a bye week. The system reprocesses the player via the pipeline to reflect the most recent player updates after these changes have been found for a particular player. These updates are accessible in the UI over a tiered Content Delivery Network, much like the batch. In order to give the user a satisfying experience, this process of regenerating changed player insights is carried out in close to real time.
In response to user requests, Watson’s Trade Assistant offers transactions. There is no caching and the computing is dynamic. As a result, Trade Assistant has 528 OpenShift pods compared to Player Insights’s 9 for on-demand computing for trade generation. With response times of under a second, we parallelize hundreds of queries per second. Other than from IBM Cloud Object Storage, the Player Insights system does not deliver any traffic from origin servers. To obtain player insights, all traffic passes via web acceleration layers.
Each project is supported by a particular set of clusters. Six worker machines with eight cores and 32GB of RAM are kept up to date by Player Insights. Trade Assistant needs 30 employees and 18 ingress nodes spread over three sites as it scales up in compute. The Akamai Global Traffic Management is used to access the tens of thousands of edge servers that make up the IBM and ESPN Content Delivery Network for Player Insights.
You’ll see in Figure 4 that the deep search method used by Trade Assistant uses a lot of cores. Trade Assistant has been designed to be as independent as possible. The data needed by the algorithm, such as club rosters and league rules, is contained in the request payloads. Different is Player Insights. All information is acquired using IBM Watson DiscoveryTM and ESPN APIs. For Player Insights, we are dependent on many external systems, such as Db2 on Cloud. The Trade Assistant on demand element does not require data saving and is independent of state.
Remember that we deployed this project concurrently with the US Open projects as well. Both made use of the same local resources. We were able to relocate and transfer containers in a portable way on Red Hat OpenShift Kubernetes as load and demand changed for each event you choose.
You can use our AI-based experiences to strengthen your squad by negotiating trades with your rivals and starting the players who have the greatest potential for improvement. Please let us know how we are doing as we get ready for the upcoming season so we can begin retraining our algorithms for you.