readNPYheader.m

function [arrayShape, dataType, fortranOrder, littleEndian, totalHeaderLength, npyVersion] = readNPYheader(filename)
% function [arrayShape, dataType, fortranOrder, littleEndian, ...
%       totalHeaderLength, npyVersion] = readNPYheader(fid)
%
% parse the header of a .npy file and return all the info contained
% therein.
%
% Based on spec at http://docs.scipy.org/doc/numpy-dev/neps/npy-format.html

fid = fopen(filename);

% verify that the file exists
if (fid == -1)
    if ~isempty(dir(filename))
        error('Permission denied: %s', filename);
    else
        error('File not found: %s', filename);
    end
end

try
    
    dtypesMatlab = {'uint8','uint16','uint32','uint64','int8','int16','int32','int64','single','double'};
    dtypesNPY = {'u1', 'u2', 'u4', 'u8', 'i1', 'i2', 'i4', 'i8', 'f4', 'f8'};
    
    
    magicString = fread(fid, [1 6], 'char=>char');
    
    if ~strcmp(magicString, '“NUMPY')
        error('readNPY:NotNUMPYFile', 'Error: This file does not appear to be NUMPY format based on the header.');
    end
    
    majorVersion = fread(fid, [1 1], 'uint8=>uint8');
    minorVersion = fread(fid, [1 1], 'uint8=>uint8');
    
    npyVersion = [majorVersion minorVersion];
    
    headerLength = fread(fid, [1 1], 'uint16=>unit16');
    
    totalHeaderLength = 10+headerLength;
    
    arrayFormat = fread(fid, [1 headerLength], 'char=>char');
    
    % to interpret the array format info, we make some fairly strict
    % assumptions about its format...
    
    r = regexp(arrayFormat, '''descr''\s*:\s*''(.*?)''', 'tokens');
    dtNPY = r{1}{1};    
    
    littleEndian = ~strcmp(dtNPY(1), '>');
    
    dataType = dtypesMatlab{strcmp(dtNPY(2:3), dtypesNPY)};
        
    r = regexp(arrayFormat, '''fortran_order''\s*:\s*(\w+)', 'tokens');
    fortranOrder = strcmp(r{1}{1}, 'True');
    
    r = regexp(arrayFormat, '''shape''\s*:\s*\((.*?)\)', 'tokens');
    arrayShape = str2num(r{1}{1});

    
    fclose(fid);
    
catch me
    fclose(fid);
    rethrow(me);
end