2. NumPy Basics

1. The n-dimensional Array (ndarray)

Why is Python slow? Because it checks the type of every single element in a loop. Why is NumPy fast? Because it cheats. It pushes the loop down to C.

Memory Layout

A Python List is an array of pointers to scattered objects. A NumPy Array is a contiguous block of memory.

• CPU Cache: When the CPU reads one number, it accidentally reads the next 16 numbers too (Cache Line).
• Locality: Because NumPy arrays are contiguous, the CPU always guesses right. This is 50x-100x faster.

2. DataTypes (Dtypes)

In Python, an integer is variable size (28 bytes+). In NumPy, you define the EXACT size.

• int8: 1 byte (-128 to 127).
• int64: 8 bytes (Standard).
• float32: 4 bytes (Single Precision for GPUs).

python
import numpy as np
# Force the array to use small integers to save RAM
a = np.array([1, 2, 3], dtype='int8')

3. Shape & Strides

How does a 1D block of memory act like a 2D matrix? Math.

• Shape: Dimensions, e.g., (3, 4) (3 rows, 4 cols).
• Strides: How many bytes to step to get to the next element/row.

If you change the shape, you just change the metadata. You do not copy the data. (Crucial for performance).