It's been a while. This last semester of school was decently overwhelming, so I took a break from maintaining the model. Sorry about that. I graduated Bucknell recently, and should be heading to UMich soon to pursue an Applied Statistics degree. This summer, I was not able to secure an internship, so I am making myself as strong of a Computer Science student as I can for the time being by learning applicable concepts such as Machine Learning.
This semester, I had the opportunity to do any project of choice for my econometrics class. I love basketball, so I used an econometric model to investigate a novel idea in the world of basketball -- using the standardized variance of the player efficiency rating to predict team winning percentage. This would prove if the balance of an NBA roster, or at least the starting 6, has an effect on a team's winningness.
I have attached the doc below if anyone cares to read. It was a lot of fun to do this project. Shoot me an email if you ever want to talk about it!
Hey guys, and welcome to a new era of sports modeling. If you followed my NFL model this past season, you're in for a treat here. I know the NBA season has started recently, so I apologize for being late with the model, but it is all ready as of now.
Instead of keeping it in Excel, I decided to make this be a lot less work for myself in the long run and coded it in Python. I have only had two classes in Python for my undergrad degree, so this took a lot of independent learning, and I was at it for a week. I thought this was a really neat project to take on, and it resembled putting together a puzzle in some way -- you have to figure out which components you need, where to get them, and how to put them together with the other pieces.
Throughout this coding experience, I picked up a decent proficiency in pandas and working with dataframes. I believe that side projects like this are really beneficial in the long run because they help create a passion for learning, and increase your technical skills significantly.
This was also my first time using Jupyter Notebook, and it was really easy and convenient to use, so if you are just starting, I highly recommend that. Along the way, I also had the opportunity to reach out to an old fraternity brother who works as a software engineer for Amazon to learn about some useful coding strategies. Shoutout to him for teaching me in his spare time.
Anyways, this model will be a bit different from the NFL model in that it won't factor in ELO, but it will heavily focus on the weighted aspects of each team's scoring, multiplying by a mystery constant that I won't reveal in here.
I learned a lot about webscraping too with this project, and it has opened my eyes to how many different datasets there are out there to work with and solve problems.
Later on in the season, I will be publishing a Tableau workbook with my results, and I will update this blog every week to talk about adjustments I make; nobody is perfect on the first try.
If you haven't checked out my NFL model from this past season, click here .The model did fairly well, hitting over 50% for spreads on 27 of the 32 teams. This was also a really cool experience that I got to increase my levels of proficiency for Excel with, and I can't wait to implement it into Python next season.
This would be something very fun to do as a career. I really liked the data wrangling aspect of this project, and the satisfaction that comes when everything in the code lines up perfectly. I learned that coding is a heavy amount of trial and error, and that perseverance is key to getting a good code out. It took many many hours of work to get these predictions, and now it's time to see if they are accurate or not.
I will try to find a way to get the model up here, even though Weebly doesn't support .py files attached to the website. Excited to see how this model turns out!
If you'd like access to the model, just write to me and I'll be happy to share!