Array Sum - Python Basics for ML

Array Sum

What you'll build

A function array_sum(arr) that returns the sum of every element of a NumPy array, using one vectorised call instead of a Python loop.

Background

Summing is the simplest reduction, and the one hiding inside most ML code: a mean loss is a sum divided by the batch size, a dot product is a sum of products, and a softmax denominator is a sum of exponentials. np.sum runs a compiled loop over contiguous memory, so it is typically about 100x faster than a Python for loop on large arrays.

Formula

array_sum(arr) = arr[0] + arr[1] + ... + arr[n-1]      (over every element, whatever the shape)

Inputs

• arr: NumPy array of numbers (integers or floats) with 1 <= arr.size <= 1000 and values in [-10^6, 10^6]. Usually 1D, but sum every element if it has more dimensions.

Expected Return Format

A single number: the total. It may be a NumPy scalar (np.int64 for integer input, np.float64 for float input); the tests compare its value.

Constraints

Use NumPy (np.sum or arr.sum()); no Python loops.

Follow-up

• How would you sum each row of a 2D array instead? Each column?
• Summing many float32 values can lose precision. Why does NumPy's pairwise summation help?

Input: array_sum(np.array([1, 2, 3, 4, 5]))
Output: 15

Input: array_sum(np.array([-1, 0, 1]))
Output: 0

Input: array_sum(np.array([0.5, 1.5, -1.0]))
Output: 1.0

import numpy as np

def array_sum(arr: np.ndarray) -> float:
    """
    Calculate the sum of all elements in a NumPy array.

    Args:
        arr: NumPy array of numbers (any shape, at least one element)

    Returns:
        The sum of all elements, as a number (a NumPy scalar is fine)
    """
    # Your code here
    pass