5 Best Ways to Create a NumPy Array of Arrays in Python

by

💡 Problem Formulation: Python developers often encounter situations where they need to create a multi-dimensional array, or an array of arrays, using the NumPy library. This task is common in scientific computing, machine learning and data analysis. For example, you might need to transform a list of lists into a structured NumPy array for efficient computation. This article explores various methods to accomplish this, with the goal of producing a NumPy array where each element is an array itself, potentially of varying sizes.

Method 1: Using numpy.array() Function

One of the most straightforward methods to create a NumPy array of arrays is utilizing the numpy.array() function. When this function is passed a list of lists (or any list-like structure), it returns a NumPy ndarray object that preserves the nested list structure, essentially creating an array of arrays if the sublists are of equal length and an array of objects if they’re not.

Here’s an example:

import numpy as np

nested_list = [[1, 2], [3, 4], [5, 6]]
numpy_array_of_arrays = np.array(nested_list)

print(numpy_array_of_arrays)

Output:

[[1 2]
 [3 4]
 [5 6]]

This code snippet first imports the NumPy library. It then defines a Python list of lists named nested_list. By passing this list to np.array(), we create a NumPy array where each sublist becomes a sub-array. The print() statement outputs a 2D structured NumPy array.

Method 2: Using numpy.vstack() Function

The numpy.vstack() function is used for vertically stacking arrays. If you have several 1D arrays and you want to stack them into a 2D array, vstack() is a powerful tool. It requires that all arrays have the same number of columns.

Here’s an example:

import numpy as np

array_one = np.array([1, 2, 3])
array_two = np.array([4, 5, 6])
array_three = np.array([7, 8, 9])

combined_array = np.vstack((array_one, array_two, array_three))

print(combined_array)

Output:

[[1 2 3]
 [4 5 6]
 [7 8 9]]

In this example, we create three 1-dimensional NumPy arrays. We then use the np.vstack() function to vertically stack these arrays, resulting in a new 2D NumPy array combined_array. The arrays are stacked in the order they are passed to vstack().

Method 3: Using numpy.concatenate() Function

numpy.concatenate() provides versatility in combining arrays. While commonly used for 1D arrays, it can concatenate multi-dimensional arrays provided they share the same shape, except for the dimension along which you’re concatenating. This makes it ideal for creating an array of arrays along a specified axis.

Here’s an example:

import numpy as np

array_one = np.array([[1, 2]])
array_two = np.array([[3, 4]])

combined_array = np.concatenate((array_one, array_two), axis=0)

print(combined_array)

Output:

[[1 2]
 [3 4]]

This snippet demonstrates the use of np.concatenate() to combine two 2D arrays along the first axis (rows). Each array is passed as an argument, and the axis=0 specifies that they should be stacked vertically, row-wise. The result is a single 2D array composed of the input arrays.

Method 4: List Comprehension with numpy.array()

List comprehension in Python, combined with the numpy.array() function, can create a NumPy array of arrays in a concise and pythonic way. This method is particularly useful when the arrays are generated based on some condition or logic defined in the comprehension.

Here’s an example:

import numpy as np

lists = [[i for i in range(j, j+3)] for j in range(1, 10, 3)]
numpy_array = np.array(lists)

print(numpy_array)

Output:

[[1 2 3]
 [4 5 6]
 [7 8 9]]

This code uses list comprehension to create a list of lists, where each inner list contains three consecutive integers. The list comprehension iterates over the numbers 1, 4, and 7 to create the starting point for each inner list. The result is converted into a NumPy array using np.array() and printed out.

Bonus One-Liner Method 5: Using numpy.arange() and reshape()

For cases where the array of arrays has a predictable structure, such as sequence-filled matrices, NumPy’s arange() function can be used in combination with reshape() method to create the array in a single line of code.

Here’s an example:

import numpy as np

numpy_array = np.arange(1, 10).reshape((3, 3))

print(numpy_array)

Output:

[[1 2 3]
 [4 5 6]
 [7 8 9]]

In this concise one-liner, np.arange(1, 10) creates a 1D array of numbers from 1 to 9. The reshape((3, 3)) method then reshapes this array into a 3×3 2D array, resulting in a perfectly structured array of arrays.

Summary/Discussion

  • Method 1: Using numpy.array(). Straightforward and versatile. Handles equal and unequal lengths of sub-arrays. Possible object dtype if sub-arrays are unequal.
  • Method 2: Using numpy.vstack(). Effective for stacking 1D arrays vertically. Requires arrays to have the same number of columns.
  • Method 3: Using numpy.concatenate(). Highly flexible for different dimensions along specific axis. Requires same shape except for the concatenation axis.
  • Method 4: List Comprehension with numpy.array(). Pythonic and useful for conditional array creations. Involves Python lists overhead.
  • Bonus Method 5: Using numpy.arange() and reshape(). Quick and efficient for patterned data. Limited to equally sized, predictable arrays.