cell2mat error in datafetch

Hi,

I have a MATLAB code that fetches data from database( n rows of 2 columns) after which cell2mat function is applied on the result set:

sql_n= strcat('Select de1, de2 from pm..pm_mast where index_type = ''IC_Scenario1''' );

IC_Sn = str2num(cell2mat(fetch(conn,sql_n)));

sn_vector = sortrows(IC_Sn,1);

Sret(:,1) =  sn_vector(:,2);

And i get the following error: CAT arguments dimensions are not consistent.

Error in ==> cell2mat at 89
          m{n} = cat(1,c{:,n});

I connect to Oracle 11g in this code. The result of the DB query has rows like this:

32 0

91 -1.6

63 0.1000375

Basically we are converting the client's code from Sybase to Oracle and this functions seems to be running fine on Sybase side.

Does something extra needs to be done on the Oracle side MATLAB code to get the cell2mat function to work properly?

[Jan]

Please format the code. Follow the "Markup help" link on this page to learn more about formatting.

Answer 1

[Jan]

1. Why do you use STRCAT if you have a single input string only? Then there is nothing to concatenate:

sql_n = strcat('Select de1, de2 from pm..pm_mast where index_type = ''IC_Scenario1''' );
% Shorter:
sql_n = 'Select de1, de2 from pm..pm_mast where index_type = ''IC_Scenario1''';

The data replied from the database as a cell do not have the same number of elements in each row. Then CELL2MAT is not able to convert them into a matrix, because a matrix must have the same number of elements in each row. So it is not a fault of CELL2MAT, but of inconsistent data.

Please inspect the value of:

fetch(conn,sql_n)

There are more efficient methods to convert a cell to a matrix than STR2NUM(CELL2MAT()).

[EDITED] If the data obtained by FETCH is a cell, which contains DOUBLE [1xN] vectors of different lengths, this might do, what you need:

C     = fetch(conn, sql_n);
Len   = cellfun('size', C, 2);
N     = numel(C);
IC_Sn = zeros(N, max(Len));  % or NaN(N, Len)?  [BUGFIX]
for i = 1:N
  IC_Sn(1:Len(i)) = C{i};
end

Then IC_Sn is a padded matrix.

[neo]

the cells do not have the same number of elements, true.

Could you suggest some better ways to do the same operation above? I will remove strcat. But should I also be doing some padding here on the Oracle side to ensure equal lengths?

de1,de2 are of type varchar2(15) and in SYBASE it is char(15). Like I mentioned, the matlab is working fine on the Sybase side.

[Jan]

I cannot suggest a solution, because this depends on your needs. If the vectors have different length a padding with zeros, Inf's or NaN's might be wanted. But it might be correct also, that the vectors have different length and storing them in a CELL is the best idea, while joining them to a matrix is too artificial.

If inserted some code to create a padded matrix - see [EDITED].

[neo]

Thanks Jan. I tried this and i have got another error that says:

Warning: Input arguments must be scalar

and this is for line:

IC_Sn = zeros(N, Len);

would you know why this is happening?

[neo]

I know that i get 144 rows of two columns each. Does this help in understanding the issue?

[Jan]

It must be "IC_Sn=zeros(N, max(Len))". A bugfix is inserted.

[neo]

I still get the same scalar error:

this is how my code looks like:

Temp_IC_Sn = fetch(conn, sql_Sn);

Len = cellfun('size', Temp_IC_Sn, 2);

N = numel(Temp_IC_Sn);

IC_Sn = zeros(N, max(Len));

for i = 1:N

IC_Sn(1:Len(i)) = Temp_IC_Sn{i};

end

ic_vector = sortrows(IC_Sn,1);

Sret(:,1) = ic_vector(:,2);