Convert numpy ndarray to matlab mlarray in python

mlarray to ndarray: np.asarray(x._data, dtype=dtype)

but inversely, ndarray to mlarray: matlab.double(x.tolist()), which is extremely slow, is there a more efficient way to do this? Thank u for answering.

Answer 1

[Al Danial]

Looks like you're making the data load unnecessarily complicated. The file actually loads cleanly into a NumPy array:

from scipy.io import loadmat
data = loadmat('data.mat', squeeze_me=True)
x = data['data']
print(x)

produces

array([[0.00000000e+00, 5.90000000e+01, 5.90000000e+01, 2.25296241e+05],
       [1.00000000e+00, 6.20000000e+01, 5.81599120e+01, 5.93159561e+04],
       [2.00000000e+00, 1.00000000e+02, 9.47518190e+01, 3.22666379e+04],
       ...,
       [2.04500000e+03, 4.00000000e+00, 4.88991300e+00, 3.01840538e+04],
       [2.04600000e+03, 2.00000000e+00, 2.26899200e+00, 6.46032757e+04],
       [2.04700000e+03, 1.00000000e+00, 1.00000000e+00, 1.18671912e+05]])

Simplify your function load_mat to

def load_mat(pth_mat, key=None):
    data = loadmat(pth_mat, squeeze_me=True)
    print(data.keys()) if key is None else None
    return data[key]

then call it like this

    x1 = load_mat('data.mat', 'data')

Answer 2

[Al Danial]

Which version of MATLAB? 2020a and newer (I don't have easy access to older versions) should just be able to do

>> mx = double(x);

without a conversion to a list.

[Gan Lee]

Thanks. Its 2021b , I tried this but not available.

[Christopher Wunder]

This is not possible for me either.

The function _is_initializer in matlab._internal.mlarray_utils.py checks for the input to be of type collections.abc.Sequence and a numpy.ndarray fails to be of such a type. The only way (without altering the package) is to convert the array beforehand or pass any kind of Sequence to it instead of an array.

Answer 3

[Al Danial]

Now I'm curious what is in your variable x. Can you make a small version of this data, write it to a .mat file, then attach the .mat file?

Answer 4

[Gan Lee]

Here is part of my python code:

# -*- coding: utf-8 -*-    

import numpy as np
from scipy.io import  loadmat
import h5py
import matlab
from matlab import engine
from matlab import mlarray


def load_mat(pth_mat, key=None):
    data = loadmat(pth_mat)
    print(data.keys()) if key is None else None
    return data.get(key)[:].astype(np.double)

def mlarray2ndarray(x: mlarray):
    return np.asarray(x._data, dtype=np.double)

def ndarray2mlarray(x: np.ndarray):
    return matlab.double(x)
    
    
if __name__ == "__main__":
    eng = engine.start_matlab()

    pth_mat = r".\data.mat"
    # np to matlab
    x1: np.ndarray = load_mat(pth_mat, key="data")
    # mx1: mlarray = ndarray2mlarray(x1) #WRONG
    mx1 = matlab.double(x.tolist()) #OK, BUT VERY SLOWLY
    
    # matlab to np
    mx2: mlarray = eng.load(pth_mat).get("data")
    x2: np.ndarray = mlarray2ndarray(mx2) #OK
    
    # ......
    
    eng.exit()
    

here is errcode:

ValueError

initializer must be a rectangular nested sequence