Why this matters
A model never sees a spreadsheet. It sees an array of numbers called features, usually written X, and an array of answers called the target, usually written y. Getting from a messy table to clean X and y is the most common task in machine learning work, and adverts that mention pipelines and data preparation are describing exactly this.
pandas for the table, NumPy for the numbers
pandas is where you load, inspect and clean. NumPy is the format models expect underneath. In practice you move from one to the other in a few lines.
Useful pandas checks before any model:
df.head()shows the first rows, so you see what the columns really contain.df.info()shows each column's type and how many values are missing.df.describe()gives counts, means and ranges, which reveals impossible values such as negative rainfall.