Train and evaluate regression models: Review

Running the second exercise, the one for training more 'powerful' models, is where this clicked. It has you run a Lasso model and then a Gradient Boosting Regressor. You just hit 'run' on mostly provided code. I'd usually complain, but it worked here. I could ignore the Python syntax and focus on the *result*. The R-squared value changed. The RMSE metric dropped. 'Model improvement' suddenly wasn't an abstract concept, but a concrete number on my screen tied to a specific action.

That's the real point of this module for someone in my role. It’s a guided tour, not a data science bootcamp in an hour. Now I have a mental model for what my team means when they say they're 'tuning hyperparameters' or 'evaluating different algorithms'. It's this: they run experiments, check the metrics, and pick the winner. The bit about using MLflow to log the runs was useful; I get it now, it's a digital lab notebook to keep track of experiments so you don't lose the best result.

An engineer would probably find this whole thing insultingly simple. But as a PM who needs to have intelligent conversations about model performance without getting bogged down in matrix algebra, it was a surprisingly productive lunch break.