Why sparse matrices save memory in SciPy - See It in Action

Why sparse matrices save memory

📖 Scenario: Imagine you have a large table of numbers where most of the values are zero. This happens often in real life, like when you track which movies people watched but most people watch only a few movies.

🎯 Goal: You will create a normal matrix and a sparse matrix with the same data. Then you will compare their memory sizes to see why sparse matrices save memory.

📋 What You'll Learn

Create a dense matrix using numpy with mostly zeros

Create a sparse matrix using scipy.sparse.csr_matrix from the dense matrix

Calculate the memory size of both matrices

Print the memory sizes to compare

💡 Why This Matters

🌍 Real World

Sparse matrices are used in recommendation systems, natural language processing, and network analysis where data is mostly zeros.

💼 Career

Understanding sparse matrices helps in optimizing memory and speed in data science and machine learning jobs.

1

Create a dense matrix with mostly zeros

Import numpy as np and create a 5x5 dense matrix called dense_matrix with these exact values: all zeros except dense_matrix[1, 2] = 10 and dense_matrix[3, 4] = 20.

SciPy

# Import numpy and create the dense matrix
# Your code here

Hint

Use np.zeros((5, 5), dtype=np.int32) to create the matrix. Then assign values to the specific positions.

2

Create a sparse matrix from the dense matrix

Import csr_matrix from scipy.sparse and create a sparse matrix called sparse_matrix from dense_matrix.

SciPy

import numpy as np

dense_matrix = np.zeros((5, 5), dtype=np.int32)
dense_matrix[1, 2] = 10
dense_matrix[3, 4] = 20

# Import csr_matrix and create sparse_matrix from dense_matrix
# Your code here

Hint

Use csr_matrix(dense_matrix) to convert the dense matrix to sparse.

3

Calculate memory sizes of both matrices

Calculate the memory size of dense_matrix using dense_matrix.nbytes and store it in dense_size. Calculate the memory size of sparse_matrix using sparse_matrix.data.nbytes and store it in sparse_size.

SciPy

import numpy as np
from scipy.sparse import csr_matrix

dense_matrix = np.zeros((5, 5), dtype=np.int32)
dense_matrix[1, 2] = 10
dense_matrix[3, 4] = 20

sparse_matrix = csr_matrix(dense_matrix)

# Calculate dense_size and sparse_size
# Your code here

Hint

Use .nbytes to get the memory size in bytes.

4

Print the memory sizes to compare

Print the text "Dense matrix size:" followed by dense_size. Then print the text "Sparse matrix size:" followed by sparse_size.

SciPy

import numpy as np
from scipy.sparse import csr_matrix

dense_matrix = np.zeros((5, 5), dtype=np.int32)
dense_matrix[1, 2] = 10
dense_matrix[3, 4] = 20

sparse_matrix = csr_matrix(dense_matrix)

dense_size = dense_matrix.nbytes
sparse_size = sparse_matrix.data.nbytes

# Print the sizes
# Your code here

Hint

Use two print statements to show the sizes with the exact text.

Practice

1. Why do sparse matrices save memory compared to regular dense matrices?

easy

A. Because they store only the non-zero elements and their positions

B. Because they compress all data into a single number

C. Because they convert numbers into strings

D. Because they store all elements including zeros explicitly

Solution

1. Step 1: Understand storage in dense matrices

   Dense matrices store every element, including zeros, which uses a lot of memory when many zeros exist.

2. Step 2: Understand storage in sparse matrices

   Sparse matrices store only non-zero values and their positions, saving memory by ignoring zeros.

3. Final Answer:

   Because they store only the non-zero elements and their positions -> Option A

4. Quick Check:

   Sparse matrices save memory by storing non-zero values only [OK]

Hint: Sparse = store only non-zero values and positions [OK]

Common Mistakes:

• Thinking sparse matrices store all zeros explicitly
• Confusing compression with sparse storage
• Believing sparse matrices convert numbers to strings

2. Which of the following is the correct way to create a sparse matrix using scipy?

easy

A. scipy.sparse.csr_matrix([[0, 0], [1, 0]])

B. scipy.sparse.dense_matrix([[0, 1], [0, 0]])

C. scipy.sparse.array([[1, 0], [0, 1]])

D. scipy.sparse.matrix([[0, 0], [0, 0]])

Solution

1. Step 1: Identify valid sparse matrix constructors in scipy

   scipy.sparse provides constructors like csr_matrix, csc_matrix, etc. 'csr_matrix' is a common sparse format.

2. Step 2: Check each option for correctness

   scipy.sparse.csr_matrix([[0, 0], [1, 0]]) uses csr_matrix correctly. Options B, C, and D use invalid or non-existent functions.

3. Final Answer:

   scipy.sparse.csr_matrix([[0, 0], [1, 0]]) -> Option A

4. Quick Check:

   Use csr_matrix to create sparse matrices [OK]

Hint: Use csr_matrix() to create sparse matrices [OK]

Common Mistakes:

• Using non-existent functions like dense_matrix or sparse.array
• Confusing sparse matrix constructors with dense matrix functions
• Forgetting to import scipy.sparse before use

3. What will be the output of this code?

import numpy as np
from scipy.sparse import csr_matrix

arr = np.array([[0, 0, 3], [4, 0, 0]])
sparse_mat = csr_matrix(arr)
print(sparse_mat.data)

medium

A. [0 0 3 4 0 0]

B. [0 3 4]

C. [3 4 0]

D. [3 4]

Solution

1. Step 1: Understand csr_matrix data attribute

   The 'data' attribute stores only the non-zero elements of the sparse matrix in row-major order.

2. Step 2: Identify non-zero elements in the array

   In arr, non-zero elements are 3 (at position [0,2]) and 4 (at position [1,0]).

3. Final Answer:

   [3 4] -> Option D

4. Quick Check:

   csr_matrix.data shows only non-zero values [OK]

Hint: csr_matrix.data shows only non-zero values [OK]

Common Mistakes:

• Expecting zeros to appear in sparse data
• Confusing data attribute with full matrix
• Misreading the order of stored elements

4. Find the error in this code that tries to create a sparse matrix:

from scipy.sparse import csr_matrix

matrix = csr_matrix([[0, 1], [2, 0]])
print(matrix.data)

medium

A. Missing import for numpy

B. csr_matrix cannot be created from a list of lists

C. No error, code runs correctly and prints non-zero values

D. csr_matrix requires dense matrix input, not list

Solution

1. Step 1: Check csr_matrix input types

   csr_matrix can accept a list of lists as input and convert it to sparse format.

2. Step 2: Verify code execution

   The code imports csr_matrix, creates a sparse matrix from a list, and prints non-zero data without error.

3. Final Answer:

   No error, code runs correctly and prints non-zero values -> Option C

4. Quick Check:

   csr_matrix accepts list input directly [OK]

Hint: csr_matrix accepts list of lists input [OK]

Common Mistakes:

• Assuming csr_matrix needs numpy arrays only
• Thinking missing numpy import causes error here
• Believing csr_matrix cannot handle lists

5. You have a large 10000x10000 matrix with 99% zeros. Which approach saves the most memory and why?

Option A: Store as a dense numpy array
Option B: Store as a scipy sparse matrix
Option C: Store as a list of lists including zeros
Option D: Store as a dictionary with all elements

hard

A. Option A, because numpy arrays are always memory efficient

B. Option B, because sparse matrices store only non-zero values and positions

C. Option C, because lists are flexible and fast

D. Option D, because dictionaries store keys and values explicitly

Solution

1. Step 1: Analyze memory use of dense numpy arrays

   Dense arrays store every element, so 10000x10000 elements use a lot of memory, including zeros.

2. Step 2: Analyze sparse matrix memory use

   Sparse matrices store only non-zero elements and their positions, greatly reducing memory when zeros dominate.

3. Step 3: Compare lists and dictionaries

   Lists of lists store all elements including zeros, using much memory. Dictionaries store keys and values, but overhead is high and all elements must be stored.

4. Final Answer:

   Option B, because sparse matrices store only non-zero values and positions -> Option B

5. Quick Check:

   Sparse matrices save memory with many zeros [OK]

Hint: Sparse matrices save memory by ignoring zeros [OK]

Common Mistakes:

• Assuming numpy arrays are always memory efficient
• Thinking lists or dictionaries save memory with many zeros
• Ignoring overhead of storing keys in dictionaries