On Wednesday, June 21, 2023, the Summer Solstice, I metaphorically look to the heavens on the longest day of the year. While staring a the sky I reflect upon the various projects past, present, and future that capture my imagination daily only to realize that I had abandoned one of my favorites despite its unfilled promise.
Now is the time to recommit to the CIELO RATINGS. The first step is to accept that the earlier versions were hindered by process and data design choices made in the early stages. Next, I must accept that my resource allocation, both temporal and fiscal, resulted in me accepting shortcuts that left me questioning the reproducibility and testability of the statistics. Finally, I must accept that the varying availability of data from different forms of racing made horse racing the worst choice to demonstrate the possibilities of extending ELO-style ratings outside of two-player contests.
For any renewed efforts working with CIELO Ratings to be worth my or the readers’ time, changes must be made. There is nothing particularly limiting to the ELO rating methodology. Unfortunately, I had previously made bad data science decisions. From the data science perspective, I didn’t tackle all the stages of the project appropriately. I was guilty of winging the previous attempt at the CIELO Ratings without a plan.
Recommitting to this project requires me to adopt what Phillip Wilkinson calls the five stages of every data science project. After defining what I wanted to accomplish by rating races with three or more contestants, I failed in the next four steps as diagramed below. Previously, I was too quick to data mine/process/model the races without rigorous thought regarding how I would continue with efficient processes.
Previous rating system attempts also trapped me in wanting to share the promise of results predictability. I often relied on anecdotal evidence to share prematurely. During this redeployment, I promised myself to be more thorough in modeling and evaluating the generated statistics before suggesting implications. Additionally, I have promised to slow the pace of production of newly derived information. Seeking validation by popular usability of the ratings also snared me into trying to create a periodic consumable piece of information whether derived odds, data visualization, or simply updating statistics on a usable frequency.
Finally, I am going to make my initial focus on a more concise data set than thoroughbred horseracing. While that is the primary arena of my career, I must embrace a simpler more concise data set to prove the value of the method while simultaneously resisting the urge to expand to all versions of racing at once.
The CIELO Ratings plan is three-pronged:
- Work on motor racing ratings in the public sphere
- Work on horse racing ratings in my private space
- Work on esoteric racing ratings only in my daydreams
Each front of the project will only progress once I have built the processes and systems to keep it sustainable. There will be no promises of when I will roll out new rating features. I hope that accompanying the metamorphosis of the project is the renewed interest of the readers in seeing a new objective form of racing ratings.
(sps – 2023-06-21)