json encode/decode

I am trying to seriaze/deserialize a model using json but a run into problem. Basically my problem is that the json encoded 1xN arrays get decoded into Nx1 arrays which compactly can be reproduced like this

>> a = [1,2,3];
>> assert( isequal(a, jsondecode(jsonencode(a))))
Assertion failed.

This was quite unexpected to me. A more expanded version of the problem goes like this

>> a = [1,2,3]
a =
1     2     3
>> s = jsonencode(a)
s =
'[1,2,3]'
>> jsondecode(s)
ans =
1
2
3

What I want is this i.e. encode/decode is reversible something like this

>> jsondecode(s)
ans = [1,2,3]

Please bear with me as I am new to MatLab but I do not underderstand this behavior. Can anyone please elaborate on this and guide me to how to obtain the result that I want? I think this is the same issue as is raised here Vector dimesions are different before encoding json and after decoding json - (mathworks.com) but I would like to avoid traversing my nested data structure and transposing all arrays.

[Jacob Sonne]

just for completeness encoding and decoding the transposed array behaves as I would expect

>> a = [1,2,3]';
>> assert( isequal(a, jsondecode(jsonencode(a))))

Likewise for a two dimensional

>> a = [1,2,3; 4,5,6];
>> assert( isequal(a, jsondecode(jsonencode(a))))

Answer 1

[Jan]

Does the encode of Matlab's production server help?

a = [1,2,3]

a = 1×3

1 2 3

j1 = jsonencode(a)

j1 = '[1,2,3]'

a1 = jsondecode(j1)

a1 = 3×1

1 2 3

j2 = mps.json.encode(a)

j2 = '[[1,2,3]]'

a2 = mps.json.decode(j2)

a2 = 1×3

1 2 3

If mps is not available on your computer, store the data in a struct and use an ugly workaround:

S.a = [1,2,3];
S   = IncludeSize(S);
j   = jsonencode(S);
S2  = jsondecode(j);

S2 = struct with fields:
        a: [3×1 double]
    size_: [1×1 struct]

S2  = RestoreSize(S2)

S2 = struct with fields:
    a: [1 2 3]

function S = IncludeSize(S)
F = fieldnames(S);
for k = 1:numel(F)
    aF   = F{k};
    data = S.(F{k});
    if isnumeric(data) % Maybe: && isrow(data)
        S.size_.(aF) = size(data);
    end
end
end

function S = RestoreSize(S)
F = fieldnames(S.size_);
for k = 1:numel(F)
    aF     = F{k};
    siz    = S.size_.(aF);
    S.(aF) = reshape(S.(aF), siz(:).');
end
S = rmfield(S, 'size_');
end

What a pity that json does not save the dimensions exhaustively.

[Jacob Sonne]

thanks @Jan. Unfortunately I don't have access to Matlab's production server mps. I might try out your tomorrow but I find it odd that the basic encode/decode cannot come back to it's original state. I can't help wondering if this is the intended behavior?

>> a = [1,2,3];
>> size(a)
ans =
1     3
>> anew = jsondecode(jsonencode(a))
anew =
1
2
3
>> size(anew)
ans =
3     1