NumPy

What is it?

NumPy introduces the ndarray (N-dimensional array), a homogeneous, memory-contiguous data structure that lets you operate on millions of numeric values without explicit Python loops. It is the foundation on which virtually the entire Python scientific stack is built: pandas, scikit-learn, SciPy, matplotlib, and TensorFlow all depend on it at their core.

What is it used for?

• Vectorized computation: apply arithmetic, logical, and aggregation operations across full arrays in a single expression — no for loops, orders of magnitude faster than native Python lists.
• Linear algebra and statistics: matrix multiplication, SVD decomposition, percentiles, means, correlations — all available in numpy.linalg and numpy.random.
• Ecosystem glue: acts as the data interchange protocol between libraries (the __array__ protocol); understanding its types (dtype) and shapes (shape) is essential for debugging issues in pandas or scikit-learn.

When to use it / when not to?

Use it when working with homogeneous numeric data: signals, images, weight matrices, simulation outputs. If you need low-level operations on raw numbers, NumPy is the right tool.

Think twice when your data is tabular with named columns of mixed types — in that case pandas or Polars provide a more expressive layer on top of NumPy without sacrificing speed. NumPy also does not replace distributed processing tools (Spark, Dask) when data does not fit in memory.

Start in 1 minute

pip install numpy

import numpy as np

# Create an array and operate on it in a vectorized way
data = np.array([1.0, 2.0, 3.0, 4.0, 5.0])

squares = data ** 2             # [1., 4., 9., 16., 25.]
mean = data.mean()              # 3.0
normalized = (data - mean) / data.std()

print(squares)
print(normalized)

Not a single for loop — every operation acts on the entire array at once. Check the official documentation for broadcasting, advanced indexing, and the full API reference.