Machine Learning is Fun

These days, I’m trying to understand why my bot accounts are getting banned. I’m betting a popular streaming service, and we often get banned 80% of our fleets. It works in the form of ban waves.

So, I’ve been trying to figure out this machine learning thing. Once you understand the process, it isn’t as scary.

The process:

  1. Get tons of data
  2. Format the data into an usable format (there are many things to do; this is called preprocessing / feature engineering; you can go without it but your results will be poorer)
  3. Train a model (the main ones are: a linear model, a Gradient Boosted Model, and obviously deep learning models)
  4. Assess the model with validation data
  5. Understand and debug your model (with e.g. SHAP)

That’s pretty much all. There is not really math involved; if you already understand Python’s (horrible) syntax, you already know machine learning.

(well, sure, there are subtleties).

If I wanted to make machine learning one of my primary goals, I would do it like this:

  1. Sign up for Toggl, create a “Learn Machine Learning” project
  2. Create a Beeminder goal of 30mn of machine learning per day
  3. Go to Kaggle.com, enter the first competition, and try to make sense of the notebooks published by others.

Here are the difficult points:

  • ML APIs look like alien speak
  • You’ll have to learn to read and use data visualisation
  • you’ll have to develop an intuition for “good data” and “bad data”

That’s pretty much it? Data-science is not really that much of a big deal. It just needs a bit of dedication. It looks less complex than hacking IMO; and with the gamifying strat’ that Kaggle’s use, I’m not worried for a shortage of data scientists… Not to mention ChatGPT that’ll be able to speak Python for us.

That’s all