Acessing dictionary from numpy savez

Question

Given two set of dict saved using numpy savez as below

import numpy as np
from numpy import load
names=['t1','t2','t3']
arr_name = np.array(names)
arr_val = np.array([1,2,3])


np.savez('data.npz', dict_one=dict(fone=arr_name,nval=arr_val),
         dict_two=dict(fone=arr_name,nval=arr_val))

When reload the data as

dict_data = load('data.npz',allow_pickle=True)
ndata=dict_data['dict_one']
opt=ndata['fone']

The compiler return an error

IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices

May I know what is the issue here,and whether this is a good practice of saving multiple dict with numpy arrays in it?

The output for print(ndata)

{'fone': array(['t1', 't2', 't3'], dtype='<U2'), 'nval': array([1, 2, 3])}

The output for print(dict_data )

<numpy.lib.npyio.NpzFile object at 0x7fe8887c6850>

Spec:

numpy 1.21.1 and Python 3.9.

Please show the full traceback. I'm unable to reproduce based on the information provided — Mad Physicist, Aug 21 '21 at 15:13
I actually reproduced this on Python 3.9.5, NumPy 1.20.3. For some reason, `type(ndata)` is `numpy.ndarray`, not `dict`... — ForceBru, Aug 21 '21 at 15:14
Did you have a stray `ndata` in the workspace before loading — Mad Physicist, Aug 21 '21 at 15:16
@balandongiv. Savez us not saving arbitrary data. Load interprets all zip entries as the binary data of an array. You are reading in garbage — Mad Physicist, Aug 21 '21 at 15:17
@balandongiv. Insufficient understanding on my part. Can you print `dict_data`? — Mad Physicist, Aug 21 '21 at 15:18
@MadPhysicist, [here's a MCVE](https://tio.run/##hZAxb8MgEIV3fgWbQUJRE88s3Tp1yWZF1inGKgoGBJdE9M87gHHUrQzo0PvevSd8wh9n@3Wdg1uodlQv3gWknwlV/Pom7Wnvi08UIrWeEAuLinLo8NiJDk/l6rsLgRDGIlGZqUN@QWIV5VV6gPmrDEdxEv2FEzJrUzwtkXGSmQgP9cuKIuikrzg6q2QZ2FymPUrYvFW27VwQup/qwaf719PyD1GpG/vgpBonQNi6GgdTqwHGuOfo9fVmlDyHu8pFCyjflqHbq@bf8EFbZJUQFJNX25wDnUdpN36u7Lq@AA "Python 3 – Try It Online") — ForceBru, Aug 21 '21 at 15:19
@balandongiv. Try `ndata=ndata.item()`. I think I can draft an answer now — Mad Physicist, Aug 21 '21 at 15:21
@ForceBru. I'm on mobile, so that's doubly appreciated. The output string was confusing me at first — Mad Physicist, Aug 21 '21 at 15:22
Does this sound like a duplicate? https://stackoverflow.com/q/17912878/2988730 — Mad Physicist, Aug 21 '21 at 15:24

score 3 · Accepted Answer · answered Aug 21 '21 at 15:22

The error says you can't index an array with a string like 'fone'.

savez saves arrays, you gave it dicts. It has wrapped each dict in the 1 element object dtype array:

In [296]: names=['t1','t2','t3']
     ...: arr_name = np.array(names)
     ...: arr_val = np.array([1,2,3])
     ...: 
     ...: 
     ...: np.savez('data.npz', dict_one=dict(fone=arr_name,nval=arr_val),
     ...:          dict_two=dict(fone=arr_name,nval=arr_val))

In [298]: dict_data = np.load('data.npz',allow_pickle=True)
     ...: ndata=dict_data['dict_one']
In [299]: ndata
Out[299]: 
array({'fone': array(['t1', 't2', 't3'], dtype='<U2'), 'nval': array([1, 2, 3])},
      dtype=object)

Use item (or indexing) to extract the element from the array

In [300]: ndata.item()
Out[300]: {'fone': array(['t1', 't2', 't3'], dtype='<U2'), 'nval': array([1, 2, 3])}
In [301]: ndata.item()['fone']
Out[301]: array(['t1', 't2', 't3'], dtype='<U2')

I think this is a duplicate but wasted my close vote: https://stackoverflow.com/q/17912878/2988730 — Mad Physicist, Aug 21 '21 at 15:25
@MadPhysicist, yes, I recently wrote a similar answer to another SO. — hpaulj, Aug 21 '21 at 15:27
An overlapping question, https://stackoverflow.com/questions/68842969/call-array-with-keyword-argument-from-npz-in-numpy. Not quite a duplicate, but part of my answer uses this `item()` trick. — hpaulj, Aug 21 '21 at 15:34

Acessing dictionary from numpy savez

1 Answers1