Convert Python tuple output into MATLAB array

Hello everyone,

I am facing a simple problem, probably due to the fact that I have not been using python for a long time.

Simply put, I wrote a function OPUSImport with Python which opens a particular file (infrared spectra from a specific company) from its FilePath. The function that allows to read this particular file read_file is available only from a Python module, brukeropusreader.

def OPUSImport(FilePath):

    from brukeropusreader import read_file

    OPUSData = read_file(FilePath)
    WaveNum = OPUSData.get_range("AB")
    Abs = OPUSData["AB"][0:len(WaveNum)]

    return WaveNum, Abs

As you can see from the function, read_file produces a Python object OPUSData and passes as output the wavenumber WaveNum (X axis) and the infrared absorbance values Abs (Y axis).

The main problem is that I would like to use this Python function into a MATLAB one:

    function Files = OPUSImport(FolderPath)
    
    Files = dir(FolderPath);
    PopLogic = true(size(Files));
    for i = 1:size(Files, 1)
        if contains(Files(i).name, ".0") % file has the correct extension
            PyOutput = py.pyOPUS.OPUSImport(FolderPath + string(Files(i).name)); % read file with the Python function, saved in the module pyOPUS
            Files(i).data.Blocks.XData = double(PyOutput(1)); 
            Files(i).data.Blocks.YData = double(PyOutput(2));
        else
            PopLogic(i) = false;
        end
    end
    Files = Files(PopLogic); % keep only structure entries with valuable data
    end

The function iteratively scans through all the data files in a folder, opens and reads the correct ones (".0"). The system works up to PyOutput, which is a tuple. The issue is that the double() function from MATLAB does not convert the tuple into a double array, with the following error:

    Error using py.tuple/double
    Python list/tuple element at position 1 must be type
    'double'.

Indeed PyOutput is a 1x2 tuple and has the following content:

    val = 
    
    
    (array([7499.95803452, 7499.92790235, 7499.89777018, ...,  370.05469869,
        370.02456652,  369.99443436]), array([0.01676818, 0.01727641, 0.01768954, ..., 0.0493812 , 0.04995102,
        0.        ]))
    
    Use string, double or cell function to convert to a MATLAB array.

Is there a better way to pass the output tuple to matlab?

Thank you in advance for the answer.

Answer 1

[Stephen23]

"The issue is that the double() function from MATLAB does not convert the tuple into a double array"

No, the issue is actually that your TUPLE contains non-scalar arrays (which are LISTs i.e. container arrays, not numeric). So rather than calling DOUBLE it would make a lot more sense to call CELL. In fact, that is exactly what MATLAB advises, when you read the message it shows (you can also read this message in your question):

T = pyrun("x = ([1,2],[98,99])","x")

T = 
  Python tuple with values:

    ([1, 2], [98, 99])

    Use string, double or cell function to convert to a MATLAB array.

DOUBLE requires that every TUPLE element is a numeric (probably scalar, but I did not check this), which your data are not (ultimately they are LISTs, not numeric):

try
    double(T)
catch ME
    ME.message
end

ans = 'Python list/tuple element at position 1 must be type 'double'.'

Solution One

Lets try CELL instead (which in general makes much more sense for a container array like TUPLE anyway):

C = cell(T)

C = 1x2 cell array
    {1x2 py.list}    {1x2 py.list}

Converting the nested LISTs is then very easy:

C = cellfun(@double,C,'uni',0)

C = 1x2 cell array
    {[1 2]}    {[98 99]}

This is also explained in the MATLAB documentation:

https://www.mathworks.com/help/matlab/matlab_external/handle-data-returned-from-python-function.html

https://www.mathworks.com/help/matlab/matlab_external/pythontuplevariables.html

https://www.mathworks.com/help/matlab/matlab_external/error-converting-list-containing-array-array.html

Or more generally:

https://www.mathworks.com/help/matlab/call-python-libraries.html

Solution Two

Another approach is to simply use the correct indexing for accessing the content of a container array (i.e. curly braces), much like what you attempted in your question (except you used parentheses, which just like every other indexing operation using parentheses returns the array itself, not its content):

V1 = double(T{1})

V1 = 1×2

1 2

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

V2 = double(T{2})

V2 = 1×2

98 99

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

Indexing into Python container types is explained in the MATLAB documentation:

https://www.mathworks.com/help/matlab/matlab_external/accessing-elements-in-python-container-types.html

Or more generally:

https://www.mathworks.com/help/matlab/matlab_external/pythontuplevariables.html

[Francesco]

Thank you @Stephen23 for your answer, this worked perfectly!

Answer 2

[Divyajyoti Nayak]

Hi @Francesco,

I see that you want a way to pass arrays in a python tuple and use them in MATLAB. The python tuple can be passed into MATLAB in a '.mat' file using the 'savemat' function from the 'SciPy' module. The '.mat' file can then be loaded into MATLAB and used as the desired MATLAB data type. Al Danial's answer in this MATLAB Answers question can help you out:

https://www.mathworks.com/matlabcentral/answers/509711-convert-a-python-tuple-containing-string-and-numerical-data-types-to-matlab?s_tid=answers_rc1-2_p2_MLT

Hope this helps!

[Francesco]

Hello @Divyajyoti Nayak,

thank you very much for your answer. Passing the information through a .mat file could be a viable option and I onestly thought about it, but I find hard to believe that there is not a better in-code way to do it. Moreover, in case of lots of data, the iterative write-read operation is time-consuming. I will implement your suggestion for now. Thank you very much!