Importing complex data from .mat file as a numpy array

Question

I have a question about tuples. I use a program which dumps large output 3-D arrays in the .mat format. They work fine in MATLAB. They are of the newer .mat format (HDF5 based). Say I generate a 10x10x10 3D matrix of complex numbers in MATLAB and save it as trial.mat

a = rand(10,10,10) + 1i*rand(10,10,10);
save('trial.mat')

Now I try to load the variable a in Python. Here is what I do.

import numpy as np
import h5py
f = h5py.File('trial.mat','r')
a = np.array(f['a'])
print a
print np.shape(a)

Here are the last few lines of the output I get.

   (0.7551184371302334, 0.15987275885464014)
   (0.5261111725119932, 0.7314707154838205)
   (0.8501109979307012, 0.05369411719045292)
   (0.8160309248196381, 0.7143895270837854)]]]
(10L, 10L, 10L)

The data is a 3D array of tuples. How do I cast this into a 3D numpy array of complex numbers? I need it for my further analysis. The actual arrays are quite large (1.5 GB and higher). I mean (0.81, 0.71) should be usable as 0.81 + 0.71j

I use MATLAB R2010a and Python 2.7.7, part of the Anaconda 2.01 (x86-64) distribution, on Windows 7.

Maybe this is related to the following question, but I am not really sure. How to read a v7.3 mat file via h5py?

Help will be greatly appreciated.

userJ · Accepted Answer · 2014-07-25 19:06:29Z

4

I'm using matlab R2010b on Windows Server 2008, and I seem to need '-v7.3' to save into an hdf5 file. Without the option, I get "version 0x0100" and with the option, I get "version 0x0200".

a = rand(10,10,10) + 1i*rand(10,10,10);
a(10,10,10) % in my case, result is 0.8667 + 0.3673i
save('trial.mat', '-v7.3')

On the Python side (in my case running under Linux): use view to make the tuples into complex values, and then repeat with real values to check that the ordering is correct:

import numpy as np
import h5py
f = h5py.File('trial.mat','r')
zz = f['a'].value.view(np.double).reshape((10,10,10,2))
zzj = zz[:,:,:,0] + 1j*zz[:,:,:,1]
zzk = f['a'].value.view(np.complex)
np.all(zzk == zzj)  # result is "True"
zzj.shape  # result is (10, 10, 10)
zzk[9,9,9].imag == f['a'][9,9,9][1]  # result is "True"
zzk[9,9,9]  # result is (0.8667498969993187+0.36728591513136899j) consistent with Matlab

We see that:

zzj and zzk have the complex values correctly ordered into real and imag
Corresponding entries into zzj and zzk match the values in Matlab

I haven't compared times with huge arrays, but I bet converting the hdf5 object directly to complex is fastest.

edited Jul 25, 2014 at 19:06

userJ

1592 silver badges8 bronze badges

answered Jul 25, 2014 at 17:41

Frank M

1,59015 silver badges16 bronze badges

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

1 Comment

userJ Over a year ago

Thanks!! Works like a charm. I had no idea there was a method to do this in numpy.

Collectives™ on Stack Overflow

Importing complex data from .mat file as a numpy array

1 Answer 1

1 Comment

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Linked

Related