source: prdatasets/prreaddataset.m @ 84

function [x,strvals] = prreaddataset(fname,strtype)
%     [X,STRVALS] = PRREADDATASET(FNAME)
%
% Read the dataset from the text file FNAME. It can process categorical
% features, or features for which categories are given in text. A matrix
% X is returned containing the numerical values, or integers. The
% integers point to the entry in STRVALS containing for each
% (categorical) feature its string members.
%
%     X = PRREADDATASET(FNAME,STRTYPE)
%
% The user can supply a vector STRTYPE that indicates for each feature
% if it is numerical (0) or string/categorical (1).
%
%     X = PRREADDATASET(FNAME,STRTYPE,DELIMITER)
%
% For datasets that have a strange delimiter (not comma or space), you
% have to supply it.
if nargin<3
        delimiter = ',';
end
if nargin<2
        strtype = [];
end

% try to open the file
[fid,message] = fopen(fname,'r');
if fid==-1
        disp(message)
        error('I cannot open file %s.',fname);
end
% get the first line:
dline = fgetl(fid);
% check if the delimiter is present:
I = find(dline==delimiter);
if isempty(I)
        delimiter = ' ';
        I = find(dline==delimiter);
        if isempty(I)
                error('Cannot determine the delimiter');
        end
end

% now run over all elements in the line:
I = [0 I length(dline)+1];
w = {};
for i=1:length(I)-1
        w{i} = dline((I(i)+1):(I(i+1)-1));
end

% remove the empty entries:
I = zeros(length(w),1);
for i=1:length(w)
        if isempty(w{i})
                I(i) = 1;
        end
end
w(find(I)) = [];
n = length(w);
x = [];

% see if we have strings or numbers, and put the result in the matrix:
strvals = {};
if isempty(strtype)
        for i=1:n
                num = str2double(w{i});
                if isnan(num)   % the feature is string
                        strtype(i) = 1;   % remember that it is a string
                        strvals{i}{1} = w{i}; % put it to the collection
                        x(1,i) = 1;
                else               % feature is a number, life is simple
                        strtype(i) = 0;
                        x(1,i) = num;
                end
        end
else
        for i=1:n
                strtype(i) = 1;   % remember that it is a string
                strvals{i}{1} = w{i}; % put it to the collection
                x(1,i) = 1;
        end
end
% now run over the other lines:
nrx = 1;
while 1
        dline = fgetl(fid);
        if ~ischar(dline), break, end  %end of file...

        % now process this line:
        nrx = nrx+1;
        % find delimiters again:
        I = find(dline==delimiter);
        % cut out the words:
        I = [0 I length(dline)+1];
        w = {};
        for i=1:length(I)-1
                w{i} = dline((I(i)+1):(I(i+1)-1));
        end
        % remove the empty entries:
        I = zeros(length(w),1);
        for i=1:length(w)
                if isempty(w{i})
                        I(i) = 1;
                end
        end
        w(find(I)) = [];
        % check:
        if length(w)~=n
                error('I cannot find enough values on line %d.',nrx);
        end
        % fill the values in the matrix
        for i=1:n
                if strtype(i)==0 % we have a number:
                        tmp = str2double(w{i});
                        if isnan(tmp)
                                error('It seems that feature %d is not numeric (encountered "%s" on line %d).',i,w{i},nrx);
                        end
                        x(nrx,i) = tmp;
                else
                        % we have to find matching strings for feature i:
                        I = strmatch(w{i},strvals{i});
                        if ~isempty(I)  % it is found
                                x(nrx,i) = I;
                        else  % we have to add this entry:
                                x(nrx,i) = length(strvals{i})+1;
                                strvals{i}{end+1} = w{i};
                        end
                end
        end
end

fclose(fid);