NumPy — The Fundamental Package for Scientific Computing

What it is

NumPy (Numerical Python) is the bedrock of the Python data science and machine learning ecosystem. It provides high-performance multidimensional arrays (ndarray), mathematical functions, linear algebra, random number generation, and Fourier transforms. Nearly every scientific Python library -- pandas, scikit-learn, TensorFlow, PyTorch -- builds on NumPy's array interface.

It is designed for scientists, data analysts, and ML engineers who need fast numerical operations in Python without writing C or Fortran code.

How it saves time or tokens

NumPy vectorized operations run 10-100x faster than equivalent Python loops because they execute in compiled C code. Broadcasting eliminates the need for explicit loops over array dimensions. The universal function (ufunc) system lets you apply operations across entire arrays in a single call.

How to use

1. Install: pip install numpy
2. Import: import numpy as np
3. Create arrays and perform operations using vectorized functions

Example

import numpy as np

# Create arrays
a = np.array([1, 2, 3, 4, 5])
b = np.random.randn(3, 4)  # 3x4 random matrix

# Basic statistics
print(f'Mean: {b.mean():.3f}, Std: {b.std():.3f}')
print(f'Matrix shape: {b.shape}')

# Linear algebra
A = np.array([[1, 2], [3, 4]])
inverse = np.linalg.inv(A)
eigenvalues = np.linalg.eigvals(A)

# Broadcasting: add a vector to each row of a matrix
matrix = np.random.randn(1000, 3)
offset = np.array([1.0, -0.5, 2.0])
result = matrix + offset  # No loop needed

# Vectorized operations: 100x faster than Python loops
x = np.linspace(0, 2 * np.pi, 10000)
y = np.sin(x) * np.exp(-x / 5)

Related on TokRepo

• AI Tools for Research -- Tools for scientific research and analysis
• Featured Workflows -- Top-rated workflows on TokRepo

Common pitfalls

• NumPy arrays have a fixed dtype; mixing integers and floats silently converts the entire array, which can cause precision issues
• Views vs copies: slicing creates a view (shared memory), not a copy; modifying a slice modifies the original array
• Large array operations can exhaust memory; use memory-mapped files (np.memmap) for datasets larger than RAM