Convert NumPy array to Python list

List() and NumPy data types

Deviate to the list() function if you want to preserve the NumPy scalar types. It keeps the NumPy types intact in the resulting list:

import numpy as np

numpy_array = np.array([np.int32(1), np.float32(2), np.complex64(3+2j)])
# Here list is the Dumbledore of Python, preserving original forms!
python_list = list(numpy_array)
print(python_list)  # Outputs: [numpy.int32(1), numpy.float32(2), numpy.complex64(3+2j)]


Exploring dimensions and shapes

Knowing the dimensions and shape of your array before converting is quite beneficial. Think of a multi-dimensional array conversion yielding a nested list resembling the shape of a Matrix inception.

Ensure data integrity

Confirm that your data transfer is exact. Double-check your NumPy array's dimensions and shape before converting:

import numpy as np

numpy_array_3d = np.array([[[1, 2], [3, 4]], [[5, 6], [7, 8]]])
print(f"Shape before conversion, almost like a pre-flight check: {numpy_array_3d.shape}")
converted_list = numpy_array_3d.tolist()
print(f"Depth of the converted list, did we get everything out?: {len(converted_list)}")


Flat layout methods

If you want to flatten your multi-level concert (array) before booing them off (conversion), you can call the ndarray.flatten() or np.ravel() methods:

import numpy as np

numpy_array_3d = np.array([[[1, 2], [3, 4]], [[5, 6], [7, 8]]])

# Using flatten() The 'I am bored, let's speed things up!' way 
flat_list = numpy_array_3d.flatten().tolist()
print(flat_list)  # Outputs: [1, 2, 3, 4, 5, 6, 7, 8]

# Also, np.ravel() works too. A fearless rival of flatten()
flat_list_ravel = np.ravel(numpy_array_3d).tolist()
print(flat_list_ravel)  # Should give the same result


More ways to Rome than one (tolist())

When .tolist() feels like a great friend but not enough to marry, alternate methods have your back.

Flattening: The B-plan

For a times when flattening needs itertools.chain.from_iterable for those 'I forgot to use NumPy' kind of days:

import numpy as np
import itertools

numpy_array_2d = np.array([[1, 2], [3, 4]])

# Bored with flatten()? Here's something different
flat_list_Bplan = list(itertools.chain.from_iterable(numpy_array_2d))
print(flat_list_Bplan)  # Outputs: [1, 2, 3, 4]


Reconversion

Had a change of heart? Turn it back into a NumPy array with np.asarray():

import numpy as np

python_list = [1, 2, 3, 4]
# Python list was fun but let's go back to NumPy array
numpy_array = np.asarray(python_list)
print(numpy_array)  # Outputs: [1, 2, 3, 4]


Shapes matter

Not happy with your array's personality (shape)? Change it:

import numpy as np

# Makeover time. Alter your array's shape as required
numpy_array = np.arange(6).reshape(2, 3)
print(numpy_array.tolist())  # Outputs: [[0, 1, 2], [3, 4, 5]]


Say Hello to comprehensions

For times when flattening arrays is cool, but not with pandas DataFrames:

import numpy as np
import pandas as pd

df = pd.DataFrame(np.array([[1, 2, 3], [4, 5, 6]]))
# The not-so-basic way of flattening
list_flattened = [element for sublist in df.values for element in sublist]
print(list_flattened)  # Outputs: [1, 2, 3, 4, 5, 6]


Post-conversion verification steps

Always run a quick health check of type, size, and shape post-conversion:

import numpy as np

numpy_array = np.array([[1, 2], [3, 4]])
converted_list = numpy_array.tolist()
print(f"Type before: {type(numpy_array)}, Size before: {numpy_array.shape}")
print(f"Type after: {type(converted_list)}, Size after: {len(converted_list)}")
