Pandas with NumPy

Pandas and NumPy interoperate constantly: columns store ndarrays, arithmetic uses broadcasting, and to_numpy() exports matrices for custom kernels or sklearn.

NumPy under the hood

import pandas as pd
import numpy as np

df = pd.DataFrame({'a': [1, 2, 3]})
arr = df['a'].to_numpy()
print(type(arr), arr.dtype, arr.shape)

Crossing the boundary

• df.to_numpy() — full 2D array (may copy if mixed dtypes)
• df['col'].values — legacy alias; prefer to_numpy()
• pd.DataFrame(arr, columns=[...]) — ndarray → labeled table
• NumPy ufuncs on Series: np.sqrt(df['x'])

Alignment caveat

NumPy ops on raw arrays ignore index labels. Pandas Series ops align on index—can introduce NaN where labels mismatch. Use .values or to_numpy() when you want pure positional NumPy behavior.

Important interview questions and answers

1. Q: to_numpy vs values?
   A: to_numpy is explicit modern API; values is legacy attribute on Series.
2. Q: Mixed dtype DataFrame?
   A: to_numpy() may upcast to object—select numeric columns first for ML.

Self-check

1. Export a numeric column to ndarray.
2. Build a DataFrame from a 2D NumPy array.

Tip: Prefer to_numpy() over legacy .values for explicit exports.