How to convert a pandas Series of lists or tuples to a Series of numpy arrays

Question

I have a csv file with x, y, and z columns that represent coordinates in a 3-dimensional space. I need to create a distance matrix from each item over all other items.

I can easily read the csv with pandas read_csv function, resulting in a DataFrame like the following:

import pandas as pd
import numpy as np

samples = pd.DataFrame(
    columns=['source', 'name', 'x', 'y', 'z'],
    data = [['a', 'apple', 1.0, 2.0, 3.0],
            ['b', 'pear', 2.0, 3.0, 4.0],
            ['c', 'tomato', 9.0, 8.0, 7.0],
            ['d', 'sandwich', 6.0, 5.0, 4.0]]
)

I can then convert the separate x, y, z columns into a Series of tuples:

samples['coord'] = samples.apply(
    lambda row: (row['x'], row['y'], row['z']),
    axis=1
)

or a Series of lists:

samples['coord'] = samples.apply(
    lambda row: [row['x'], row['y'], row['z']],
    axis=1
)

But I cannot create a Series of arrays:

samples['coord'] = samples.apply(
    lambda row: np.array([row['x'], row['y'], row['z']]),
    axis=1
)

I get the ValueError, "Shape of passed values is (4,3), indices imply (4,6)"

I'd really like to have the data prepped so that I can simply call the scipy's distance_matrix function, which expects two arrays, as follows:

dmat = scipy.spatial.distance_matrix(
    samples['coord'].values,
    samples['coord'].values
)

I am, of course, open to any more pythonic or more efficient way to achieve this goal if my approach is poor.

Answer 1

This stores NumPy array in coords :

samples['coord'] = list(samples[['x', 'y', 'z']].values)

Now:

>>> samples.coord[0]
array([ 1.,  2.,  3.])

Answer 2

I figured out that I can just extract a numpy array from the dataframe and use it to get the distance matrix.

sample_array = np.array(samples[['x', 'y', 'z']])
dmat = scipy.spatial.distance_matrix(sample_array, sample_array)

But I'd still like to have those little arrays embedded in the dataframe, alongside the other data, and I'd upvote and accept an answer that can do that.