Distance computation (distance.cdist) in SciPy

Introduction

We use distance computation to find how far apart points are from each other. It helps us compare data points in many tasks like clustering or searching.

Comparing locations of stores to find the closest ones.

Grouping similar customers based on their shopping habits.

Finding nearest neighbors in recommendation systems.

Measuring similarity between images or text features.

Syntax

SciPy

scipy.spatial.distance.cdist(XA, XB, metric='euclidean', *args, **kwargs)

XA and XB are arrays of points (rows are points, columns are features).

metric defines how distance is measured, default is 'euclidean' (straight line distance).

Examples

Compute Euclidean distance between two sets of 2D points.

SciPy

from scipy.spatial import distance
import numpy as np

XA = np.array([[0, 0], [1, 1]])
XB = np.array([[1, 0], [2, 2]])

result = distance.cdist(XA, XB, metric='euclidean')
print(result)

Compute Manhattan (city block) distance instead of Euclidean.

SciPy

result = distance.cdist(XA, XB, metric='cityblock')
print(result)

Compute cosine distance to measure angle difference between points.

SciPy

result = distance.cdist(XA, XB, metric='cosine')
print(result)

Sample Program

This program calculates the straight-line distances between each point in points_A and each point in points_B. The result is a matrix where each row corresponds to a point in points_A and each column corresponds to a point in points_B.

SciPy

from scipy.spatial import distance
import numpy as np

# Define two sets of points
points_A = np.array([[0, 0], [3, 4]])
points_B = np.array([[1, 1], [6, 8]])

# Compute Euclidean distances between each point in A and each point in B
distances = distance.cdist(points_A, points_B, metric='euclidean')

print('Distance matrix:')
print(distances)

Important Notes

The output is a 2D array where element (i, j) is the distance between XA[i] and XB[j].

You can use many distance metrics like 'euclidean', 'cityblock', 'cosine', 'hamming', etc.

Input arrays must be 2D with shape (number_of_points, number_of_features).

Summary

cdist computes distances between two groups of points.

It returns a matrix showing distances from each point in the first group to each point in the second.

You can choose different ways to measure distance using the metric parameter.

Practice

1.

What does the scipy.spatial.distance.cdist function do?

easy

A. Generates random points within a specified range.

B. Calculates the sum of all points in a single array.

C. Computes distances between two sets of points and returns a matrix of distances.

D. Sorts points in ascending order based on their coordinates.

Solution

1. Step 1: Understand the function purpose

   cdist calculates distances between two groups of points, not sums or sorting.

2. Step 2: Identify the output type

   The function returns a matrix showing distances from each point in the first group to each point in the second.

3. Final Answer:

   Computes distances between two sets of points and returns a matrix of distances. -> Option C

4. Quick Check:

   cdist = distance matrix [OK]

Hint: Remember: cdist = cross distances between two point sets [OK]

Common Mistakes:

• Thinking cdist sums points instead of distances
• Confusing cdist with sorting or random generation
• Assuming cdist returns a single distance value

2.

Which of the following is the correct way to import and use cdist from scipy.spatial.distance?

?

easy

A. from scipy import cdist
result = cdist(A, B, metric='euclidean')

B. import scipy.distance.cdist
result = cdist(A, B, metric='euclidean')

C. import cdist from scipy.spatial.distance
result = cdist(A, B, metric='euclidean')

D. from scipy.spatial.distance import cdist
result = cdist(A, B, metric='euclidean')

Solution

1. Step 1: Check correct import syntax

   The correct import is from scipy.spatial.distance import cdist. Other forms are invalid or cause errors.

2. Step 2: Confirm usage syntax

   Calling cdist(A, B, metric='euclidean') is correct to compute Euclidean distances.

3. Final Answer:

   from scipy.spatial.distance import cdist result = cdist(A, B, metric='euclidean') -> Option D

4. Quick Check:

   Correct import and call = from scipy.spatial.distance import cdist result = cdist(A, B, metric='euclidean') [OK]

Hint: Use 'from scipy.spatial.distance import cdist' to import [OK]

Common Mistakes:

• Using incorrect import statements causing ImportError
• Trying to import cdist directly from scipy
• Using wrong syntax like 'import cdist from ...'

3.

Given the code below, what is the output?

import numpy as np
from scipy.spatial.distance import cdist

A = np.array([[0, 0], [1, 1]])
B = np.array([[1, 0], [2, 2]])
result = cdist(A, B, metric='euclidean')
print(result)

medium

A. [[1. 2.82842712]
[1. 1.41421356]]

B. [[1.41421356 2.23606798]
[1. 1.41421356]]

C. [[1. 1.41421356]
[1.41421356 2.23606798]]

D. [[0. 1.41421356]
[1. 2.23606798]]

Solution

1. Step 1: Calculate distances from points in A to points in B

   Distances: - From [0,0] to [1,0] = sqrt((1-0)^2 + (0-0)^2) = 1.0 - From [0,0] to [2,2] = sqrt((2-0)^2 + (2-0)^2) ≈ 2.8284 - From [1,1] to [1,0] = sqrt((1-1)^2 + (0-1)^2) = 1.0 - From [1,1] to [2,2] = sqrt((2-1)^2 + (2-1)^2) ≈ 1.4142

2. Step 2: Form the distance matrix

   Matrix rows correspond to points in A, columns to points in B: [[1.0, 2.8284], [1.0, 1.4142]]

3. Final Answer:

   [[1. 2.82842712] [1. 1.41421356]] -> Option A

4. Quick Check:

   Euclidean distances matrix = [[1. 2.82842712] [1. 1.41421356]] [OK]

Hint: Calculate each pair distance carefully, row by row [OK]

Common Mistakes:

• Mixing row and column order in the output matrix
• Using wrong distance formula (e.g., Manhattan instead of Euclidean)
• Rounding errors causing confusion

4.

What is wrong with the following code snippet?

import numpy as np
from scipy.spatial.distance import cdist

A = np.array([[0, 0], [1, 1]])
B = np.array([1, 0])
result = cdist(A, B, metric='euclidean')
print(result)

medium

A. The metric 'euclidean' is not supported by cdist.

B. B should be a 2D array, but it is 1D, causing a shape error.

C. cdist requires both inputs to be lists, not numpy arrays.

D. The import statement for cdist is incorrect.

Solution

1. Step 1: Check input array shapes

   cdist expects both inputs to be 2D arrays representing sets of points. Here, B is 1D, which is invalid.

2. Step 2: Understand error cause

   Passing 1D array B causes a shape mismatch error because cdist tries to interpret rows as points.

3. Final Answer:

   B should be a 2D array, but it is 1D, causing a shape error. -> Option B

4. Quick Check:

   Input arrays must be 2D for cdist [OK]

Hint: Always use 2D arrays for both inputs in cdist [OK]

Common Mistakes:

• Passing 1D arrays instead of 2D arrays
• Assuming cdist accepts lists only
• Using unsupported metric names

5.

You have two sets of points representing locations of stores and customers:

stores = np.array([[0, 0], [5, 5], [10, 10]])
customers = np.array([[1, 1], [6, 6], [8, 8], [12, 12]])

Using cdist, how can you find the closest store for each customer?

hard

A. Compute the distance matrix with cdist(customers, stores) and find the index of the minimum distance in each row.

B. Compute the distance matrix with cdist(stores, customers) and find the index of the maximum distance in each column.

C. Use cdist(stores, customers) and sum all distances to find closest stores.

D. Calculate Euclidean distance manually for each pair without using cdist.

Solution

1. Step 1: Compute distance matrix from customers to stores

   Use cdist(customers, stores) to get distances where each row is a customer and each column is a store.

2. Step 2: Find closest store for each customer

   For each customer (row), find the index of the minimum distance value. This index corresponds to the closest store.

3. Final Answer:

   Compute the distance matrix with cdist(customers, stores) and find the index of the minimum distance in each row. -> Option A

4. Quick Check:

   Closest store = min distance per customer row [OK]

Hint: cdist(customers, stores), then argmin per row finds closest store [OK]

Common Mistakes:

• Swapping order of inputs causing wrong distance matrix shape
• Using max instead of min to find closest
• Summing distances instead of finding minimum