Exploring the numpy triu Method in Python

Rate this post

πŸ’‘ Problem Formulation: When dealing with matrices in Python, it’s often necessary to extract the upper triangular part. For example, if we have a square matrix, we may want to isolate the upper triangle, including the diagonal, which contains elements (i,j) where i ≀ j. The numpy.triu method provides a simple way to achieve this. Below, we explore various ways to utilize this method effectively.

Method 1: Using numpy triu with Default Parameters

The simplest use of the numpy.triu method involves passing just the matrix from which you want to extract the upper triangle. By default, numpy.triu includes the main diagonal and all elements above it, effectively setting the lower triangle to zero. This method is straightforward and requires minimal code.

Here’s an example:

import numpy as np

matrix = np.array([[1, 2, 3],
                   [4, 5, 6],
                   [7, 8, 9]])
upper_triangle_matrix = np.triu(matrix)


The output of this code snippet is:

[[1 2 3]
 [0 5 6]
 [0 0 9]]

This code snippet demonstrates the creation of a 3×3 matrix and the application of the numpy.triu function. The output is the original matrix but with all the elements below the main diagonal set to zero, thus revealing the upper triangular matrix.

Method 2: Specifying the k-th Diagonal with numpy triu

Numpy’s triu method can also take an optional parameter k that allows you to define which diagonal and above to include. Setting k=0 includes the main diagonal, while k=1 would exclude the main diagonal and include only elements above it.

Here’s an example:

upper_triangle_matrix_k1 = np.triu(matrix, k=1)


The output of this code snippet is:

[[0 2 3]
 [0 0 6]
 [0 0 0]]

In this code snippet, the k=1 parameter is used to exclude the main diagonal. The resulting output includes only elements above the main diagonal.

Method 3: Working with Non-Square Matrices

The numpy.triu method is not limited to square matrices. It can be applied to any two-dimensional numpy array to retrieve the upper triangular elements. Elements that are not part of the upper triangle due to the shape of the matrix will simply be left unchanged.

Here’s an example:

rectangular_matrix = np.array([[1, 2, 3, 4],
                               [5, 6, 7, 8]])
upper_triangle_rectangular = np.triu(rectangular_matrix)


The output of this code snippet is:

[[1 2 3 4]
 [0 6 7 8]]

This example illustrates the application of the numpy.triu function on a non-square (2×4) matrix. As expected, the operation zeroes out the elements in the lower triangle based on the conceptual extension of the main diagonal.

Method 4: Generating a Triangular Matrix Directly

Instead of starting with a full matrix and extracting the upper triangle, numpy can create a triangular matrix directly using numpy.tri to generate a mask, and then multiplying it element-wise with the numpy.multiply method or the asterisk (*) operator.

Here’s an example:

tri_mask = np.tri(*matrix.shape, k=0, dtype=bool)
upper_triangle_direct = np.multiply(matrix, tri_mask)


The output of this code snippet is:

[[1 0 0]
 [4 5 0]
 [7 8 9]]

This snippet first creates a boolean triangular mask using numpy.tri. Multiplication of the original matrix with this mask directly yields the lower triangular matrix.

Bonus One-Liner Method 5: Using Comprehension and Slicing

For a Pythonic and compact approach, one could create an upper triangular matrix using a one-liner list comprehension, although this method may not be as efficient or readable as the numpy.triu function.

Here’s an example:

upper_triangle_comp = np.array([row[:i+1]+[0]*(len(row)-i-1) for i, row in enumerate(matrix)])


The output of this code snippet is:

[[1 2 3]
 [0 5 6]
 [0 0 9]]

This example uses list comprehension to iterate through the rows of the matrix and slices to construct the upper triangle explicitly. It requires knowledge of Python’s list slicing and comprehension syntax but achieves the same result as numpy.triu.


  • Method 1: Default usage of numpy triu. Strength: Simple and easy to use. Weakness: None.
  • Method 2: Specifying k-th diagonal. Strength: Flexibility in choosing the diagonal. Weakness: Requires understanding of k parameter.
  • Method 3: Non-square matrices handling. Strength: Versatility with different shapes. Weakness: Can be less intuitive for non-square cases.
  • Method 4: Generating matrix directly. Strength: Direct approach, potentially more efficient. Weakness: Involves additional steps and concepts such as masking.
  • Method 5: One-liner list comprehension. Strength: Pythonic and compact. Weakness: Potentially less efficient and readable, not utilizing numpy’s optimizations.