1

I am trying to create a file to be read in a matlab enviroment. The structure in matlab looks like this

trx(1) = 
          x: [1×1500 double]
          y: [1×1500 double]
          a: [1×1500 double]
          b: [1×1500 double]
      theta: [1×1500 double]
 firstframe: 1
   endframe: 1500
    nframes: 1500
        off: 0 


 trx(2) = 
          x: [1×751 double]
          y: [1×751 double]
          a: [1×751 double]
          b: [1×751 double]
      theta: [1×751 double]
 firstframe: 750
   endframe: 1500
    nframes: 751
        off: -749

So naturally I created a python dictionary with the required fields and create a list, then used savemat. However when I loaded in matlab I only get cell arrays. I also tried using this but the problem is that not all of the fields are arrays with the same shapes for example 'firstframe' is an int. Then when I used fromarrays() but it complains because the shape does not match.

I am trying now to convert a dictionary to an structured array, but have not found anything related. And also trying to create a numpy record that allows different shapes for the arrays. Any light very welcome

Ivan Felipe Rodríguez
  • 189
  • 16
  • 1
    The mapping between Python objects and MATLAB is tricky. You might try creating a sample in MATLAB, save it, and then `loadmat`. That should give a clearer idea of what MATLAB needs. – hpaulj Feb 08 '19 at 18:42
  • I'm a little rusty on this, but I think MATLAB matrix maps onto a order F `ndarray`. A `cell` onto an object dtype array. A `struct` on to a structured array (i.e. one with a compound `dtype`, record array in your link). The arrays that go into a structured array do have to match in shape. – hpaulj Feb 08 '19 at 18:43

1 Answers1

2

In Octave

M =

  scalar structure containing the fields:

    x =

       1   2   3   4

    y =

       5   6   7   8

    one =  1
    two =

       1   2

>> save -7 struct.mat M

In Ipython:

In [450]: dat = io.loadmat('struct.mat')
In [451]: dat
Out[451]: 
{'__header__': b'MATLAB 5.0 MAT-file, written by Octave 4.2.2, 2019-02-08 18:49:49 UTC',
 '__version__': '1.0',
 '__globals__': [],
 'M': array([[(array([[1., 2., 3., 4.]]), array([[5., 6., 7., 8.]]), array([[1.]]), array([[1., 2.]]))]],
       dtype=[('x', 'O'), ('y', 'O'), ('one', 'O'), ('two', 'O')])}

Here M is (1,1) structured array, with all fields being object dtype. That way they can each have their own shape. The scalar is a (1,1) matrix.

hpaulj
  • 221,503
  • 14
  • 230
  • 353
  • Hey hpaulji your answer really moved me closer. Now I am stuck with this problem . When I try to recreate the object : Setting void-array with object members using buffer. – Ivan Felipe Rodríguez Feb 08 '19 at 21:24