Saving and loading Python dict with savemat results in error

Question

Here is an minimal example of the error I get. If I understood the documentation correctly, this should be working, but it seems I did not.

a={}
a['test1']=1
a['test2']=2
a['test3']=3
import scipy.io as io
io.savemat('temp',{'a':a})
b = io.loadmat('temp')
b['a'].keys()

Traceback (most recent call last):
  File "<input>", line 1, in <module>
AttributeError: 'numpy.ndarray' object has no attribute 'keys'

Answer 1

You seem to be operating under the assumption that scipy.io.savemat is intended to be able to save a standard dictionary. I don't believe that is the case. The dictionary argument holds the names of numpy arrays which are written out into the Matlab file. So you can do something like this

import scipy.io as io
import numpy as np

y1=np.array([1,2,3,4])
y2=np.array([10,20,30,40])
y3=np.array([100,200,300,400])

a={}
a['test1']=y1
a['test2']=y2
a['test3']=y3
io.savemat('temp',a)
b = io.loadmat('temp')

print b['test1']
print b['test2']
print b['test3']

which gives:

[[1]
 [2]
 [3]
 [4]]
[[10]
 [20]
 [30]
 [40]]
[[100]
 [200]
 [300]
 [400]]

Answer 2

It looks like loadmat returns recarray instead of dict. I checked with scipy 0.9.0. Equivalent of b['a'].keys() will be b['a'].dtype.names.

Examples:

In [12]: b['a'].shape
Out[13]: (1, 1)

In [14]: b['a'].dtype.names
Out[16]: ('test1', 'test3', 'test2')

In [17]: b['a']['test1']
Out[17]: array([[[[1]]]], dtype=object)