Zhaitan Zhaitan - 4 months ago 16
Python Question

Trouble using numpy.load

I have the following data, written in python 2, that I'd like to load to a python 3 file.

import numpy as np
x = np.array([{'a': np.array([1., 2., 3])}])
np.save('data.npy', x)


My first attempt was this:

import numpy as np
x = np.load('data.npy')

UnicodeError: Unpickling a python object failed


After playing around with the original data I am trying to load, it seems that whenever I have a numpy float inside of a numpy array inside a python dictionary, I get the error. I can load a dictionary, I can load a numpy array, I can even load a numpy array inside a python dictionary, but as soon as I have numpy floats inside a numpy array inside a python dictionary, I get the error. My second attempt was this:

import numpy as np
x = np.load('data.npy', encoding = 'bytes')
x

array([{b'a': array([ 1., 2., 3.])}], dtype=object)


This worked in that I could load the data, but it added a 'b' in front of every key in the dictionary. I was wondering if anyone had any insight into why this problem is occuring and how to fix it.

Thanks!

edit:

it seems like the following solved the problem:

import numpy as np
x = np.load('data.npy', encoding = 'latin1')


I'm not quite sure why, but it seemed to work.

Answer

The default encoding in Python 2 is ascii; in Python 3 it is utf-8. latin1 (a.k.a., ISO-8859-1) is a superset of ascii. That's why loading ascii-encoded strings with latin1 works and gives the same result as loading it with ascii.

Comments