Import matlab cell array into python for scikit-learn

Question

I have a 1x81 cell array in matlab.

Each cell is a 30x30 matrix of doubles.

I want to store this in python (for use in scikit-learn) with the shape (81,30,30).

I've read a few questions here and worked through their code but I'm not having any success.

you need to reshape your cell array to a matrix then stackoverflow.com/questions/37620330/… — GameOfThrows
– GameOfThrows, Commented Nov 1, 2016 at 15:36
I would suggest that you look at this other question too: stackoverflow.com/questions/1761419/… — Eskapp
– Eskapp, Commented Nov 1, 2016 at 16:25

piman314 · Accepted Answer · 2016-11-01 17:19:33Z

5

You can do this just using scipy.io.loadmat. But you have to be careful because of some of the differences in the formats.

from scipy import io
import numpy as np

C = io.loadmat('test.mat')
print type(C)
print C.keys()

Outputs:

<type 'dict'>
['C', '__version__', '__header__', '__globals__']

So you can see that scipy is inlcuding a bunch more information that we don't really need, but we can see your cell C.

C = C['C']
print type(C)

Ouputs:

<type 'numpy.ndarray'>

Okay so that's got use the Cell from Matlab.

print C.shape

Ouputs:

(1, 81)

Which isn't quite right, but with a bit of processing we can get it the way you want.

C = np.squeeze(C)
X = np.empty((C.shape[0], C[0].shape[0], C[0].shape[1]))
for i in xrange(X.shape[0]):
    X[i] = C[i]
print X.shape

Outputs:

(81, 30, 30)

Voila, we have your cell in a numpy array. Just as a forward warning, in general scikit-learn takes a 2D array as an input, not a 3D array.

answered Nov 1, 2016 at 17:19

piman314

5,35526 silver badges36 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Thank you - that works perfectly! Thanks for explaining it through each step