My Story in Becoming a Bayesian

Before Sunrise

Back in university, I had no interest in the statistics I studied at that time. When I found out that I have a course on probability and statistics I was like yes new mathematical topics. But I was still not attached. There is a team project for the course that the professor asked us to do, and professor, if you’re reading this, I owe you, you sparked all of my desires now. He gave us two options; a report on Markov Decision Process (MDP) and analysing the elements used in a model that the professor has built for one of his research papers, and a common topic related to building a data model which I can’t recall its details. I was the team leader and a teammate and I convinced the team to choose the first one. The time was way prior to the AI blooming. It’s a team effort, obviously, and my role was to flourish that effort, so I gave my heart to make this happen. We read a lot of resources ourselves to comprehend the topic, write it, and read the paper again, and again, and again to understand how the elements were used. I personally came across Markov chains during that research, and here, my dear reader, is when the spark happened.

Before Sunset

Never heard of the topic again, but from time to time I revisited the report because it’s one, if not the utmost, report or project that I’ve enjoyed doing till today. Later, in my first job, I was asked to find a way to analyze the customers’ behaviour, yes, if you know, that’s where I found Markov chain, again. To further analyse them in details, I read about advanced statistical techniques including Markov Chain Monte Carlo (MCMC). Bayesian statistics started to appear everywhere. Still not much invested as I was more into how to use it.

Before Midnight

To analyse the tactics and external factors that drive sales or revenue, there’s a common statistical data analysis method called Marketing Mix Modeling (MMM). It turned out to the team (my teammates, at that time, and I) that the most explainable library used that is not a black box and can illustrate its output is pymc-marketing library, developed by PyMC Labs. Its reliability to me is based on the team’s analysis on the papers that’s based on, and the benchmarks provided by the developers. Please take that with a pinch of salt. I’ve seen how powerful Bayesian statistics is, at least mathematically, and how it could be used to build the superior products whatever the purpose is.

Before Tomorrow

Bayesian statistics is a way of figuring things out by updating what you already believe with new information. You start with an initial guess or baseline fact (called a prior), look at new data or evidence (called the likelihood), and combine them to get a revised, more accurate conclusion (called the posterior). Think of it like guessing how likely it is to rain today: you start with yesterday’s weather report, look out the window to see dark clouds forming, and combine both facts to decide if you should carry an umbrella. It seems trustworthy to you, right?

Please, let your mind think for a moment and brainstorm what other applications there could be. Imagine them! They are all doable!

I basically became Bayesian and updated my belief. Whatever I do, I’ll make sure to bring clarity and transparency to the mystery. That also could be updated based on whatever life holds for me- the new information.