Documentation of detrend_NaN2

Global Index (short | long) | Local contents | Local Index (short | long)

Function Synopsis

y = detrend_NaN(xdat, tol, show);

Help text

  function Y = detrend_NaN ( Xdat , tol , show ) ;

   Y = detrend(Xdat) removes the best linear fit to
   each column of Xdat.  If Xdat is N-dimensional, then it is
   assumed that the time series Xtim will be removed from the first
   dimension of Xdat.  Missing points are ignored, and the data are
   assumed to be uniformly spaced in time.

   tol is the number of points required for a trend to be computed on an
   individual column of xdat:  tol <= size(find(~isnan(xdat(:,i)))).

   show = 'show', 'noshow', or 1, 0, respectively.

   This routine is quicker than detrend_NaN and produces the same
   result, within round-off error.

Listing of function detrend_NaN2

function y = detrend_NaN(xdat, tol, show);

sz = size(xdat); ndim = length(sz);
if (ndim == 2) & (sz(1) == 1); xdat = xdat(:); end;
sz = size(xdat); ndim = length(sz);

if nargin < 3; tol = 2; end;

if nargin < 4; show = 0; end;
if isstr(show);
  show = strcmp(show, 'show');
end

[m1, n1] = size(xdat);

xtim = [[1:m1]' ones(m1, 1)];
[m2, n2] = size(xtim);

%  Condense data to get rid of NaN's in Xtim
if n2 > 1;
  kp_time = find(~isnan(sum(xtim'))); 
else
  kp_time = find(~isnan(xtim));
end
nkp_time = length(kp_time);
xdat = xdat(kp_time,:); xtim = xtim(kp_time,:);

%  Start regressions and removals
if show; disp(['Number of iterations:  ' num2str(n1)]); end;

c = repmat(NaN, [nkp_time, n1]);
for i = 1:n1;
  if show; 
    disp(['Iteration ' num2str(i) ' of ' num2str(n1)]); 
  end;
  if n2 > 1;
    kp = find(sum( ~isnan((xdat(:,i)*ones(1,n2)) .* xtim)') == n2 );
  else
    kp = find(~isnan(xdat(:,i) .* xtim));
  end
  nkp = length(kp);
  if nkp > tol
    c(kp,i) = xdat(kp,i) - xtim(kp,:)*(xtim(kp,:)\xdat(kp,i));
  end
end

%  Reshape output so it is the same dimension as input

y = repmat(NaN, [m1, n1]);
y(kp_time,:) = c;

if ndim > 2;
  y = reshape(y, sz);
end