Every model you'll ever train stands on three tools: NumPy (arrays instead of loops), Pandas (tables with judgment), and Matplotlib/Seaborn (plots that tell the truth). This phase converts you from "writes loops over lists" to thinks in shapes and columns — pandas and NumPy fluency built in on the way.
Pick a Kaggle dataset you actually care about (your finlytics expenses, network traffic,
stock history). One notebook, publishable: load → clean (missing values, dtypes, documented
decisions) → explore (groupbys, correlations) → 6+ honest plots → a written summary of three
findings someone else could verify. This becomes your EDA template for every dataset in Phases 3–7.
Full spec + dataset ideas in the phase guide ↗ · finish the free Kaggle Pandas cert along the way.