CSR format (Compressed Sparse Row) in SciPy - Step-by-Step Execution

Concept Flow - CSR format (Compressed Sparse Row)

Start with Sparse Matrix

↓

Extract non-zero values

↓

Record column indices of non-zeros

↓

Build row pointer array

↓

Store as CSR format: data, indices, indptr

CSR format stores a sparse matrix by saving only non-zero values, their column positions, and pointers to row starts.

Execution Sample

SciPy

import numpy as np
from scipy.sparse import csr_matrix

matrix = np.array([[0,0,1],[2,0,0],[0,3,0]])
csr = csr_matrix(matrix)
print(csr.data)
print(csr.indices)
print(csr.indptr)

This code converts a dense matrix to CSR format and prints its internal arrays.

Execution Table

Step	Action	Data Array	Indices Array	Indptr Array
1	Start with matrix [[0,0,1],[2,0,0],[0,3,0]]
2	Extract non-zero values	[1, 2, 3]
3	Record column indices of non-zeros	[1, 2, 3]	[2, 0, 1]
4	Build row pointer array	[1, 2, 3]	[2, 0, 1]	[0, 1, 2, 3]
5	Store as CSR format	[1, 2, 3]	[2, 0, 1]	[0, 1, 2, 3]

💡 CSR arrays fully built representing the sparse matrix.

Variable Tracker

Variable	Start	After Step 2	After Step 3	After Step 4	Final
data	[]	[1, 2, 3]	[1, 2, 3]	[1, 2, 3]	[1, 2, 3]
indices	[]	[]	[2, 0, 1]	[2, 0, 1]	[2, 0, 1]
indptr	[0]	[0]	[0]	[0, 1, 2, 3]	[0, 1, 2, 3]

Key Moments - 2 Insights

Why does the 'indices' array have values [2, 0, 1] instead of sorted?

The 'indices' array records the column positions of non-zero values in the order they appear row-wise, matching the original matrix layout (see execution_table step 3).

What does the 'indptr' array represent?

'indptr' shows where each row's data starts in the 'data' array. For example, indptr[1]=1 means row 1's data starts at data[1] (see execution_table step 4).

Visual Quiz - 3 Questions

Look at the execution_table at step 3, what is the value of indices[1]?

A 2

B 1

C 0

D 3

💡 Hint

Check the 'indices' array column in execution_table row for step 3.

At which step does the 'indptr' array get its final form?

A Step 2

B Step 4

C Step 3

D Step 5

💡 Hint

Look at the 'indptr' column in execution_table and see when it changes from [0] to [0,1,2,3].

If the matrix had an extra zero row at the end, how would 'indptr' change?

A [0, 1, 2, 3, 3]

B [0, 1, 2, 3]

C [0, 1, 2, 3, 4]

D [0, 1, 2, 4]

💡 Hint

The last number in 'indptr' equals total non-zero elements; zero row adds no new data but adds a pointer.

Concept Snapshot

CSR format stores sparse matrices efficiently by:
- data: non-zero values in row order
- indices: column indices of these values
- indptr: pointers to row starts in data
Use csr_matrix() in scipy to convert dense to CSR
Access arrays via .data, .indices, .indptr

Full Transcript

CSR format compresses sparse matrices by storing only non-zero values, their column positions, and row start pointers. The example matrix [[0,0,1],[2,0,0],[0,3,0]] has non-zero values [1,2,3]. Their column indices are [2,0,1] respectively. The indptr array [0,1,2,3] marks where each row's data starts in the data array. This structure saves memory and speeds up matrix operations. The execution table shows step-by-step how these arrays build up. Key points include understanding that indices follow the order of non-zero elements row-wise and indptr length is number of rows plus one. Quizzes test understanding of these arrays and their changes.