Mercurial > hg > ltpda
view m-toolbox/classes/@ao/melementOp.m @ 52:daf4eab1a51e database-connection-manager tip
Fix. Default password should be [] not an empty string
author | Daniele Nicolodi <nicolodi@science.unitn.it> |
---|---|
date | Wed, 07 Dec 2011 17:29:47 +0100 |
parents | f0afece42f48 |
children |
line wrap: on
line source
% MELEMENTOP applies the given matrix operator to the data. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % DESCRIPTION: MELEMENTOP applies the given matrix operator to the data. % % CALL: % a = melementOp(callerIsMethod, op, opname, opsym, minfo, pl, a1, a2,...) % % % VERSION: $Id: melementOp.m,v 1.12 2011/04/18 16:55:43 ingo Exp $ % % HISTORY: 01-02-07 M Hewitson % Creation % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function varargout = melementOp(varargin) import utils.const.* % Settings callerIsMethod = varargin{1}; op = varargin{2}; opname = varargin{3}; opsym = varargin{4}; % Info to pass to history iobj = varargin{5}; pl = varargin{6}; % variable names varnames = varargin{8}; % Collect AO inputs but preserve the element shapes % ... also collect numeric terms and preserve input names argsin = varargin{7}; args = {}; in_names = {}; for kk=1:numel(argsin) if isa(argsin{kk}, 'ao') args = [args argsin(kk)]; in_names = [in_names varnames(kk)]; elseif isnumeric(argsin{kk}) % When promoting the number to an AO, we have to be sure to call % the fromVals and allow it to add history. a = fromVals(ao, plist('vals', argsin{kk}), 0); args = [args {a}]; if all(size(argsin{kk}) == [1 1]) in_names = [in_names num2str(argsin{kk})]; elseif any(size(argsin{kk}) == [1 1]) in_names = [in_names 'vector']; else in_names = [in_names 'matrix']; end end end if numel(args) < 2 error('### %s operator requires at least two AO inputs.', opname) end if numel(args) == 2 % get the two arrays a1 = args{1}; a2 = args{2}; % check the data for kk=1:numel(a1) if ~isa(a1(kk).data, 'ltpda_data') error('### one of the input AOs has an empty data field'); end end for kk=1:numel(a2) if ~isa(a2(kk).data, 'ltpda_data') error('### one of the input AOs has an empty data field'); end end % Here we operate on two AO arrays according to the rules %---------- Deal with error cases first r1 = size(a1,1); c1 = size(a1,2); r2 = size(a2,1); c2 = size(a2,2); %== Rule 4: [1xN] */ [Nx1] if r1 == 1 && r2 == 1 && c1==c2 && c1>1 error('### It is not possible to %s two AO vectors of size [1xN]', opname); end %== Rule 6: [Nx1] */ [Nx1] if r1 == r2 && c1==1 && c2==1 && r1>1 error('### It is not possible to %s two AO vectors of the size [Nx1]', opname); end %== Rule 7: [NxP] */ [Nx1] if r1 == r2 && c1>1 && c2==1 && c1~=r1 && r1>1 error('### It is not possible to %s [NxP] and [Nx1]', opname); end %== Rule 8: [NxP] */ [Px1] if c1 == c2 && r1>1 && r2==1 && c1>1 error('### It is not possible to %s [NxP] and [1xP]', opname); end %== Rule 9: [NxP] */ [NxP] if isequal(size(a1), size(a2)) && r1>1 && c1>1 if size(a1,1) ~= size(a1,2) error('### It is not possible to %s [NxP] and [NxP]', opname); end end %------------- Now perform operation if numel(a1)==1 || numel(a2)==1 % Rules 1,2,5 if isvector(a1) || isvector(a2) || ismatrix(a1) || ismatrix(a2) % Rule 2,5: vector or matrix + single AO if isvector(a1) || ismatrix(a1) res = copy(a1,1); for ee=1:numel(res) res(ee).data = compatibleData(res(ee),a2); res(ee).data.setY(operate(a1(ee), a2)); res(ee).data.setDy(operateError(a1(ee), a2)); % set history and name if ~callerIsMethod names = getNames(in_names, res(ee), ee, a2, []); res(ee).addHistory(iobj, pl, names(1:2), [res(ee).hist a2.hist]); res(ee).name = names{3}; end res(ee).data.setYunits(getYunits(a1(ee), a2)); end else res = copy(a2,1); for ee=1:numel(res) res(ee).data = compatibleData(res(ee),a1); res(ee).data.setY(operate(a2(ee), a1)); res(ee).data.setDy(operateError(a2(ee), a1)); % set history and name if ~callerIsMethod names = getNames(in_names, a1, [], res(ee), ee); res(ee).addHistory(iobj, pl, names(1:2), [a1.hist res(ee).hist]); res(ee).name = names{3}; end res(ee).data.setYunits(getYunits(a1, a2(ee))); end end else % Rule 1: [1x1] */ [1x1] res = copy(a1,1); res.data = compatibleData(res,a2); res.data.setY(operate(a1, a2)); res.data.setDy(operateError(a1, a2)); % set history and name if ~callerIsMethod names = getNames(in_names, res, [], a2, []); res.addHistory(iobj, pl, names(1:2), [res.hist a2.hist]); res.name = names{3}; end res.data.setYunits(getYunits(a1, a2)); end elseif isvector(a1) && isvector(a2) && r1==1 && c2==1 && r2==c1 % Rule 3: [1xN] */ [Nx1] if strcmp(op, 'mrdivide') error('### It is not possible to divide two matrices with different sizes'); end res = []; if strcmp(op, 'mtimes') inner = 'times'; else inner = 'rdivide'; end for ee=1:numel(a1) if isempty(res) res = feval(inner,a1(ee),a2(ee)); else res = res + feval(inner,a1(ee),a2(ee)); end end elseif isvector(a1) && isvector(a2) && r1>1 && c1==1 && r2==1 && c2>1 % Rule 5: [Nx1] */ [1xM] res(r1,c2) = ao(); for kk=1:r1 for ll=1:c2 res(kk,ll) = feval(op,a1(kk),a2(ll)); end end elseif ismatrix(a1) && (ismatrix(a2) || isvector(a2)) if strcmp(op, 'mrdivide') && ~isequal(size(a1),size(a2)) error('### Can only divide matrices of the same size'); end % Rule 10: matrix */ matrix res(r1,c2) = ao; for kk=1:r1 for ll=1:c2 res(kk,ll) = feval(op,a1(kk,:),a2(:,ll)); end end else error('### The inputs were not properly handled. This shouldn''t happen.'); end % Did something go wrong? if isempty(res) error('### The inputs were not properly handled. This shouldn''t happen.'); end else % we recursively pass back to this method res = copy(args{1}, 1); for kk=2:numel(args) res = feval(op, res, args{kk}); end end % Set output varargout{1} = res; %---------- nested functions %------------------------------------------------- % Check the two inputs have compatible data types function dout = compatibleData(a1,a2) %== Data types if (isa(a1.data, 'fsdata') && isa(a2.data, 'tsdata')) || ... isa(a2.data, 'fsdata') && isa(a1.data, 'tsdata') error('### Can not %s time-series data to frequency-series data.', opname); end % check X units for all data types if ~isa(a1.data, 'cdata') && ~isa(a2.data, 'cdata') if ~isempty(a1.data.xunits.strs) && ~isempty(a2.data.xunits.strs) if a1.data.xunits ~= a2.data.xunits error('### X units should be equal for the %s operator', op); end end end % determine output data type d1 = copy(a1.data,1); d2 = copy(a2.data,1); if isa(d1, 'data2D') && isa(d2, 'data2D') if numel(d1.y) > 1 dout = d1; elseif numel(d2.y) > 1 dout = d2; else dout = d1; end elseif isa(d1, 'data2D') && isa(d2, 'cdata') dout = d1; elseif isa(d1, 'cdata') && isa(d2, 'data2D') dout = d2; else dout = d1; end end function uo = getYunits(a1, a2) % For other operators we need to apply the operator uo = feval(op, a1.data.yunits, a2.data.yunits); end % Perform the desired operation on the data function y = operate(a1, a2) y = feval(op, a1.data.y, a2.data.y); end % Perform the desired operation on the data uncertainty function dy = operateError(a1, a2) if ~isempty(a1.dy) || ~isempty(a2.dy) da1 = a1.dy; da2 = a2.dy; if isempty(da1) da1 = zeros(size(a1.y)); end if isempty(da2) da2 = zeros(size(a2.y)); end switch op case {'plus', 'minus'} dy = sqrt(da1.^2 + da2.^2); case {'times', 'mtimes'} dy = sqrt( (da1./a1.y).^2 + (da2./a2.y).^2 ) .* abs(a1.y.*a2.y); case {'rdivide', 'mrdivide'} dy = sqrt( (da1./a1.y).^2 + (da2./a2.y).^2 ) .* abs(a1.y./a2.y); otherwise dy = []; end else dy = []; end end %----------------------------------------------- % Get two new AO names from the input var names, % the input AO names, and the indices. function names = getNames(in_names, a1, jj, a2, kk) % First variable name if isempty(a1.name) && ~isempty(in_names{1}) if ~isempty(jj) if numel(jj) == 1 names{1} = sprintf('%s(%d)', in_names{1}, jj); else names{1} = sprintf('%s(%d,%d)', in_names{1}, jj(1), jj(2)); end else names{1} = in_names{1}; end else if ~isempty(jj) if numel(jj) == 1 % names{1} = sprintf('%s(%d)', a1.name, jj); names{1} = sprintf('%s', a1.name); else % names{1} = sprintf('%s(%d,%d)', a1.name, jj(1), jj(2)); names{1} = sprintf('%s', a1.name); end else names{1} = a1.name; end end % Second variable name if isempty(a2.name) && ~isempty(in_names{2}) if isempty(in_names{2}) in_names{2} = a2.name; end if ~isempty(kk) if numel(kk) == 1 names{2} = sprintf('%s(%d)', in_names{2}, kk); else names{1} = sprintf('%s(%d,%d)', in_names{2}, kk(1), kk(2)); end else names{2} = in_names{2}; end else names{2} = a2.name; if ~isempty(kk) if numel(kk) == 1 % names{2} = sprintf('%s(%d)', a2.name, kk); names{2} = sprintf('%s', a2.name); else % names{2} = sprintf('%s(%d,%d)', a2.name, kk(1), kk(2)); names{2} = sprintf('%s', a2.name); end else names{2} = a2.name; end end % The output AO name names{3} = sprintf('(%s%s%s)', names{1}, opsym, names{2}); end %------------------------------------- % Return true if the input is a matrix function r = ismatrix(a) if nrows(a) > 1 && ncols(a) > 1 r = true; else r = false; end end %------------------------------------- % Return true if the input is a vector function r = isvector(a) if (nrows(a)==1 && ncols(a)>1) || (ncols(a)==1 && nrows(a)>1) r = true; else r = false; end end %------------------------------------- % Return numnber of rows in the array function r = nrows(a) r = size(a,1); end %------------------------------------- % Return numnber of cols in the array function r = ncols(a) r = size(a,2); end end % End of add % END