From lists to arrays

A NumPy array (numpy.ndarray) stores a fixed type of number in one contiguous block of memory — no per-element type-checking, no scattered objects to chase down. This is exactly the same idea that lets a spreadsheet compute a column sum instantly even with thousands of rows: uniform data laid out predictably is much faster to process in bulk than a mixed bag of arbitrary objects.

import numpy as np

scores_list = [88, 92, 79, 95, 61]        # plain Python list
scores = np.array(scores_list)             # NumPy array

print(type(scores), scores.dtype)          # ndarray, int64 (or similar)

The dtype (data type) is worth noticing — it's new information a plain list doesn't have. Every element in scores is the same type, which is precisely the constraint that unlocks speed.

| | Python list | NumPy array | |---|---|---| | Can mix types | Yes | No — one dtype per array | | Resize freely | Yes | Awkward — arrays are usually recreated, not resized | | Math across all elements | Manual loop | Built-in, vectorized (next page) | | Speed on large numeric data | Slow | Fast |

Neither is "better" in general — a list of mixed-type records (a name, a score, a flag) is still a perfectly reasonable Python list. NumPy arrays earn their keep specifically for numeric data you're going to compute across, which describes almost everything in machine learning: scores, pixel values, embeddings, weights.

Every model you'll train later — a decision tree, a neural network, anything in scikit-learn or PyTorch — expects its input as arrays (or pandas structures built on top of arrays), not plain Python lists. Getting comfortable with arrays now means the data-handling code in every future lab looks familiar rather than like new syntax to decode on top of a new algorithm.

Real datasets aren't just a single row of numbers — they're often a grid: many examples (rows), each with several features (columns). NumPy represents this as a 2D array, and its .shape attribute tells you the grid's dimensions directly:

grid = np.array([[88, 92], [79, 95], [61, 70]])
print(grid.shape)   # (3, 2) — 3 rows, 2 columns

.shape becomes one of the first things you check whenever something goes wrong in a data pipeline — a surprising number of bugs in real ML code are simply a shape mismatch between what one step produced and what the next step expected.