Inconsistent type following cell2table

Converting cell array with cell2table results in inconsistent conversion of arrays depending on (as far as I can tell) other members of the table column. Unfortunately this is causing issues for me, is there any simple fix that does not require changing inputs on a case-by-case basis?

MRE:

test = {[1, 2], [nan]; [nan], [nan]; [nan], [nan]};

test_table = cell2table(test);

test_table{2, 1}{1};  % okay (returns nan)
test_table{2, 2}{1};  % error: Brace indexing is not supported for variables of this type.

[Stephen23]

"Converting cell array with cell2table results in inconsistent conversion of arrays depending on (as far as I can tell) other members of the table column."

The CELL2TABLE documentation states here how columns are handled depending on their array sizes:

• "If the contents of the cells in a column of C have compatible sizes and types, then the corresponding table variable is the vertical concatenation of those contents into an array."
• "If the contents of the cells in a column have different sizes and types, then the corresponding table variable is a cell array."

Your 1st column will be a cell array (because [1,2] cannot be concatenated vertically with NaN), and because scalar NaNs can be concatenated the 2nd column will be numeric. Lets now check with your example data:

C = {[1,2],nan;nan,nan;nan,nan}

C = 3×2 cell array
    {[1 2]}    {[NaN]}
    {[NaN]}    {[NaN]}
    {[NaN]}    {[NaN]}

T = cell2table(C)

T = 3×2 table
      C1       C2 
    _______    ___

    {[1 2]}    NaN
    {[NaN]}    NaN
    {[NaN]}    NaN

So far eveything seems to be working exactly as documented and as expected.

What do you expect the 1st column to be?

What do you expect the 2nd column to be?

[JZ]

Thanks for your answer and pointing me to the relevant part of the documentation, I missed that.

Good to know it is working as expected, although I think this is a design decision that should be revisited. I do not think it is good practice to have a function that casts the same datatype to two different datatypes within the same table. I would have no particular expectation for them to be [nan] or nan after conversion, but I would expect consistency.

Answer 1

[Star Strider]

Use parentheses, not curly braces, to index into a vector within a cell —

test = {[1, 2], [nan]; [nan], [nan]; [nan], [nan]};

test_table = cell2table(test)

test_table = 3×2 table
     test1     test2
    _______    _____

    {[1 2]}     NaN 
    {[NaN]}     NaN 
    {[NaN]}     NaN 

test_table{2, 1}(1)   % okay (returns nan)

ans = 1×1 cell array
    {[NaN]}

test_table{2, 2}(1)   % error: Brace indexing is not supported for variables of this type.

ans = NaN

test_table{1,1}

ans = 1×1 cell array
    {[1 2]}

test_table{1,1}{1}(1)

ans = 1

test_table{1,1}{1}(2)

ans = 2

The indexing can get a bit complicated.

.

[JZ]

Thanks for your answer, unfortunately this results in the same problem that the datatype is cast differnently across the table, but I see that is the expected behaviour. I can check the type later in the code, thanks for the workaround!

[Star Strider]

As always, my pleasure!

[Stephen23]

@JZ: it seems that you could avoid the different column conversions of CELL2TABLE by... not using CELL2TABLE. Why not just access the cell array itself? Then your indexing would always be the same, no errors, correct data.