Mercurial > octave
changeset 24415:3cd47be4fcbe
fplot.m: Fix handling of non-vectorized input functions (bug #52673).
* fplot.m: Change documentation to note that performance is better with
vectorized functions, but that scalar functions are accepted. If function is
determined to be scalar during input validation, use arrayfun to convert it to
a vectorized version. Add eps to denominator of calculation of error so that 0
values do not result in NaN. Add new BIST test for warning emitted when
function is not vectorized.
| author | Rik <rik@octave.org> |
|---|---|
| date | Fri, 15 Dec 2017 09:21:26 -0800 |
| parents | 2adff1743db7 |
| children | c341ebb40ede |
| files | scripts/plot/draw/fplot.m |
| diffstat | 1 files changed, 18 insertions(+), 11 deletions(-) [+] |
line wrap: on
line diff
--- a/scripts/plot/draw/fplot.m Thu Dec 14 21:23:39 2017 -0800 +++ b/scripts/plot/draw/fplot.m Fri Dec 15 09:21:26 2017 -0800 @@ -66,10 +66,11 @@ ## discontinuities are unlikely to plot well. This restriction may be removed ## in the future. ## -## @code{fplot} requires that the function accept and return a vector argument. -## Consider this when writing user-defined functions and use @code{.*}, -## @code{./}, etc. See the function @code{vectorize} for potentially -## converting inline or anonymous functions to vectorized versions. +## @code{fplot} performance is better when the function accepts and returns a +## vector argument. Consider this when writing user-defined functions and use +## element-by-element operators such as @code{.*}, @code{./}, etc. See the +## function @code{vectorize} for potentially converting inline or anonymous +## functions to vectorized versions. ## ## @seealso{ezplot, plot, vectorize} ## @end deftypefn @@ -138,16 +139,19 @@ y = feval (fn, x); endif + if (isscalar (y0)) + warning ("fplot: FN is not a vectorized function which reduces performance"); + fn = @(x) arrayfun (fn, x); # Create a new fn that accepts vectors + y0 = feval (fn, x0); + y = feval (fn, x); + endif + if (rows (x0) == rows (y0)) fcn_transpose = false; elseif (rows (x0) == columns (y0)) fcn_transpose = true; y0 = y0.'; y = y.'; - elseif (isscalar (y0)) - ## FN is a constant value function - y0 = repmat (y0, size (x0)); - y = repmat (y, size (x)); else error ("fplot: invalid function FN (# of outputs not equal to inputs)"); endif @@ -165,7 +169,7 @@ yi = interp1 (x0, y0, x, "linear"); ## relative error calculation using average of [yi,y] as reference ## since neither estimate is known a priori to be better than the other. - err = 0.5 * max (abs ((yi - y) ./ (yi + y))(:)); + err = 0.5 * max (abs ((yi - y) ./ (yi + y + eps))(:)); if (err < tol || abs (err - err0) < tol/2) ## Either relative tolerance has been met OR ## algorithm has stopped making any reasonable progress per iteration. @@ -237,11 +241,11 @@ %!test %! ## Constant value function -%! fn = @(x) 5; +%! fn = @(x) 0; %! [x, y] = fplot (fn, [-1, 1]); %! assert (columns (y) == 1); %! assert (rows (x) == rows (y)); -%! assert (y, repmat ([5], size (x))); +%! assert (y, repmat ([0], size (x))); ## Test input validation %!error fplot (1) @@ -254,3 +258,6 @@ %!error <invalid function FN> %! fn = @(x) [x;x]; %! fplot (fn, [-1,1]); +%!warning <FN is not a vectorized function> +%! fn = @(x) 0; +%! [x,y] = fplot (fn, [-1,1]);