function [zzgrid, xvec] = ffndgrid(x, f, delta,limits,aver);
%FFNDGRID Fast 'n' Furious N-D data gridding.
%
% CALL: [fgrid, xvec] = ffndgrid(x,f, delta, limits, aver );
%
% fgrid = Matrix of gridded data.
% xvec = cellarray of gridvectors
% xl1:dx1:xu1 or linspace(xl1,xu1,Nx1) depending on delta.
% x = [x1 x2,...xD] coordinate matrix for unevenly spaced data, f.
% size NxD.
% f = f(x), vector of function values length N.
% delta = [dx1+i*pad, dx2 ,...,dxD] or [-Nx1+i*pad -Nx2,...,NxD], where
% dx1 to dxD and Nx1 to NxD defines the stepsize of grids and
% number of bins, respectively, in the D dimesional space. Empty
% gridpoints are padded with IMAG(delta(1))=pad. (default -75)
% limits = [xl1 xu1 xl2 ...xuN fl fu n0], contain the limits in the
% D-dimensional x1-x2...xN-f-plane of where to grid. The last
% parameter, n0, removes outliers from the data set by ignoring
% grid points with n0 or less observations. When n0 is negative
% it is treated as the percentage of the total number of data points.
% (default [min(x1),max(x1),min(x2),.... max(xN),min(f),max(f),0])
% aver = 0 If each value of fgrid is the sum of f values for all points falling
% within each cell.
% 1 If each value of fgrid is the average of f values for all points falling
% within each cell. (default)
% 2 If each value of fgrid is the minimum of f values for all points falling
% within each cell. (default)
% 3 If each value of fgrid is the maximum of f values for all points falling
% within each cell. (default)
%
% FFNDGRID grids unevenly spaced D-dimensinoal data in vector f into a matrix fgrid.
%
% NOTE: - The vector limits can be padded with NaNs if only
% certain limits are desired, e g if xl1 and fl are wanted:
%
% ffndgrid(x, f, [],[.5 nan nan nan 45])
%
% - If no output arguments are given, FFGRID will plot the gridded
% function with the prescribed axes using PCOLOR (for
% D=2 only).
%
% Examples:
% N = 500;D=2; sz = [N ,D ];
% x = randn(sz); z = ones(sz(1),1);
% [nc, xv] = ffndgrid(x,z,-15,[],0); % Histogram
% pcolor(xv{:},nc) %
% [XV,YV]=meshgrid(xv{:});
% text(XV(:),YV(:),int2str(nc(:)))
% dx = [diff(xv{1}(1:2)) diff(xv{2}(1:2))];
% contourf(xv{:}, nc/(N*prod(dx))) % 2-D probability density plot.
% colorbar
% colormap jet
%
% See also: griddata
%
%
% Tested on: MatLab 4.2, 5.0, 5.1, 5.2, 5.3, 6.5, 7 and 7.01.
% History:
% revised pab 16.09.2004
% - added aver options
% - way of gridding modified (while instead of sparse)
% - Sparse matrix if density (rho) < 0.4 & D>3.
% - Dimensional permutation for D=3 removed (if was a bug for full matrices).
% modified by R.BREGEON - Physicist, PHD - FRANCE
%
% revised pab 02.08.2001
% - made it general for D dimensions + changed name from ffgrid to ffndgrid
% -added nargchk + examples.
% -updated help header to wafo-style
% - moved dx and dy into delta =[dx,dy]
% -removed call to bin
% modified by Per A. Brodtkorb
% 05.10.98 secret option: aver
% optionally do not take average of values for each point
% 12.06-98
% by
% 28.7.97, Oyvind.Breivik@gfi.uib.no.
%
% Oyvind Breivik
% Department of Geophysics
% University of Bergen
% NORWAY
error(nargchk(2,5,nargin))
[r, c] = size(x);
if r==1,% Make sure x is a column vector.
x = x(:);
end
[N,D]=size(x);
f = f(:);
if length(f)==1, f = f(ones(N,1),:) ;
elseif length(f)~=N,error('The length of f must equal size(x,1)!'),end
% Default values
dx = repmat(-75,1,D);
pad = 0;
xyz = [zeros(1,2*D) min(f), max(f), 0];
xyz(1:2:2*D) = min(x,[],1);
xyz(2:2:2*D) = max(x,[],1);
% take average of values for each point (default)
if (nargin < 5)|isempty(aver), aver = 1; end
if (nargin >= 4) & ~isempty(limits),
nlim = length(limits);
ind = find(~isnan(limits(1:min(7,nlim))));
if any(ind), xyz(ind) = limits(ind);end
end
if nargin>=3&~isempty(delta),
Nd =length(delta);
delta = delta(1:min(Nd,D));
if Nd ==1, delta = repmat(delta,1,D);end
ind = find(~(isnan(delta)|delta==0));
if any(ind),
dx(ind) = real(delta(ind));
pad = imag(delta(1));
end
end
xL = xyz(1:2:2*D);
xU = xyz(2:2:2*D);
fL = xyz(end-2);
fU = xyz(end-1);
n0 = xyz(end);
ind = find(dx<0);
if any(ind),
if any(dx(ind)~=round(dx(ind))),
error('Some of Nx1,...NxD in delta are not an integer!'),
end
dx(ind) = (xU(ind)-xL(ind))./(abs(dx(ind))-1);
end
% bin data in D-dimensional-space
binx = round((x - xL(ones(N,1),:))./dx(ones(N,1),:)) +1;
fgsiz = ones(1,max(D,2));
xvec = cell(1,D);
for iy=1:D,
xvec{iy} = xL(iy):dx(iy):xU(iy);
fgsiz(iy) = length(xvec{iy});
end
if D>1
in = all((binx >= 1) & (binx <= fgsiz(ones(N,1),:)),2) & (fL <= f) & (f <= fU);
else
in = (binx >= 1) & (binx <= fgsiz(1)) & (fL <= f) & (f <= fU);
end
binx = binx(in,:);
f = f(in);
N = length(binx); % how many datapoints are left now?
Nf = prod(fgsiz);
if D>1,
fact = [1 cumprod(fgsiz(1:D-1))];
binx = sum((binx-1).*fact(ones(N,1),:),2)+1; % linear index to fgrid
end
% Fast gridding
% fgrid = sparse(binx,1,f,Nf,1);% z-coordinate
ix=binx;
% ******* PART to customize *******
switch 1
case aver==0 | aver==1 % Mean value or density map case for f
while any(ix)
ib=find(ix~=0,1);
vl=(ix==ix(ib));
fgrid(ix(ib),1)=sum(f(vl));
ix(vl)=0;
end
case aver==2 % minima map for f
while any(ix)
ib=find(ix~=0,1);
vl=(ix==ix(ib));
fgrid(ix(ib),1)=min(f(vl));
ix(vl)=0;
end
case aver==3 % maxima map for f
while any(ix)
ib=find(ix~=0,1);
vl=(ix==ix(ib));
fgrid(ix(ib),1)=max(f(vl));
ix(vl)=0;
end
% ===== ADD your own case here for customize (aver=4, aver=5,....) ==========
otherwise
disp('error')
return
end
% ******** END of part to customize
fgrid(end+1:Nf)=0;
if n0~=0 | aver==1
ngrid = sparse(binx,1,ones(N,1),Nf, 1); % no. of obs per grid cell
if(n0 < 0), n0 = -n0*N; end % n0 is given as percentage
if n0~=0,
% Remove outliers
fgrid(ngrid <= n0) = 0;
ngrid(ngrid <= n0) = 0;
N = full(sum(ngrid(:))); % how many datapoints are left now?
end
end
ind = find(fgrid); % find nonzero values
if aver==1
fgrid(ind) = fgrid(ind)./ngrid(ind); % Make average of values for each point
end
if pad~=0,
Nil=find(fgrid==0);
fgrid(Nil) = fgrid(Nil)+pad; % Empty grid points are set to pad
end
rho = length(ind)/(Nf); % density of nonzero values in the grid
if rho<=0.4 & D>3
fgrid = sparse(fgrid);
end
if r==1
fgrid = fgrid.';
else
fgrid = reshape(fgrid,fgsiz);
if D==2
fgrid=fgrid.';
end
end
if (nargout > 0)
zzgrid = fgrid;
elseif D==2,% no output, then plot
colormap(flipud(hot)) %colormap jet
if 1,
%figure('Position', [100 100 size(fgrid)])
imagesc(xvec{:}, fgrid)
set(gca,'YDir','normal')
else
pcolor(xvec{:}, fgrid)
shading flat %interp
end
colorbar
xlabel(inputname(1))
ylabel(inputname(1))
zstr=inputname(2);
dum = size(fgrid');
if (~isempty(zstr)) % all this vital information ...
str = sprintf('Color scale: %s, %d data points, grid: %dx%d, density: %4.2f', ...
inputname(3), N, dum(1), dum(2), rho);
else
str = sprintf('%d data points, grid: %dx%d, density: %4.2f', ...
N, dum(1), dum(2), rho);
end
title(str);
end
