bootstrp: Fail safe to guarantee that the sum of the weights equals n * nboot

Author: acp29

inst/bootstrp.m

@@ -364,19 +364,10 @@
     w = cellfun (@(n) ones (n, 1) / n, n, 'UniformOutput', false);
     s = num2cell (ones (1, nvar), 1);
   else
-    if (isnumeric (w))
-      w = repmat (mat2cell (w, n{1}, 1), 1, nvar);
-    end
-    if (any (arrayfun (@(v) any (bsxfun (@lt, w{v}, 0)), 1 : nvar)))
-      error ('bootstrp: Weights cannot contain negative values')
-    end
-    if (any (arrayfun (@(v) any (isinf(w{v})), 1 : nvar)))
-      error ('bootstrp: Weights cannot contain any infinite values')
-    end
-    if (any (arrayfun (@(v) any (isnan(w{v})), 1 : nvar)))
-      error ('bootstrp: Weights cannot contain NaN values')
-    end
     if (match)
+      if (isnumeric (w))
+        w = repmat (mat2cell (w, n{1}, 1), 1, nvar);
+      end
       if (numel (w) > 1)
         if (~ isequal (w{:}))
           error (cat (2, 'bootstrp: Weights must be the same for each row', ...
@@ -385,15 +376,31 @@
       end
     else
       if (~ all (bsxfun (@eq, cat (2, 1, nvar), size (w))))
-        error (cat (2, 'bootstrp: Weights must be an array of cells equal ', ...
-                       ' equal in number to their non-matching data arguments'))
+        error (cat (2, 'bootstrp: Weights must be an array of cells equal', ...
+                       ' in number to their non-matching data arguments'))
       end
     end
+    if (any (arrayfun (@(v) any (bsxfun (@lt, w{v}, 0)), 1 : nvar)))
+      error ('bootstrp: Weights cannot contain negative values')
+    end
+    if (any (arrayfun (@(v) any (isinf(w{v})), 1 : nvar)))
+      error ('bootstrp: Weights cannot contain any infinite values')
+    end
+    if (any (arrayfun (@(v) any (isnan(w{v})), 1 : nvar)))
+      error ('bootstrp: Weights cannot contain NaN values')
+    end
     s = arrayfun (@(v) fzero (@(s) sum (round (s * w{v} / mean (w{v}) * nboot) ...
                                 - nboot), 1), 1 : nvar, 'UniformOutput', false);
   end
   w = arrayfun (@(v) round (s{v} * w{v} / mean (w{v}) * nboot), ...
                             1 : nvar, 'UniformOutput', false);
+  if (~ isempty (w))
+    % Fail safe to guarantee that the sum of the weights equals n * nboot
+    for v = 1:nvar
+      [wmax, imax] = max (w{v});
+      w{v}(imax) = wmax + nboot * n{v} - sum (w{v});
+    end
+  end
 
   % Perform balanced bootstrap resampling
   if (match)