Thanks again Eric,
Your examples are exactly what I was after.
My colleague was hypothesizing that there's probably a less-than instead of a less-than-or-equal somewhere, if it is a bug.
---- Eric Firing <efiring@...202...> wrote:
> Thanks Eric.
> However, when I specify the same number of levels as suggested, contourf divides this example into three regions, with a diagonal 'stripe' instead of a clean boundary, so I guess I'm asking whether it's possible to trick contourf into generating a single boundary between the two regions such that it matches that found by contour?
Now I see the problem; this is something of a corner case, but it may be
pointing to a bug.
Where do you really want the line to fall?
Do you need to specify the number of contours instead of specifying the
actual levels (boundaries)? Are you actually dealing with zeros and
ones as in the example? If so, you probably want
contourf(a, [-1, 0.5, 2], colors=('w', 'k'), extend='neither')
contourf(a, [0.5, 2], colors=('k',), extend='neither')
In this case you are saying "color everything between 0.5 and 2, and
Specifying one contour instead of giving the levels is yielding 0.6;
this is a consequence of using the MaxNLocator by default to auto-select
> For the moment, a suitable workaround seems to be to do
> where the background colour is white. This generates what I'm after.
> I notice also that linewidths is mentioned in the docstring under Obsolete:, but it seems to do nothing, so it should probably be removed from the docstring.
I will fix the docstring. Thanks.
> thanks again,
> ---- Eric Firing <efiring@...202...> wrote:
>> Gary Ruben wrote:
>>> I'm notice that contourf behaves differently to contour by default in
>>> where it decides to position contours. For example, using pylab, if you try
>>> I'd have expected the contours to line up, but they don't. Is there a
>>> way to get contourf to place its contours at the same position as contour?
>> Specify the same number of levels:
>> That takes care of this simple case. There are other cases, however,
>> where contour and contourf simply don't agree; contouring is ambiguous,
>> and only part of the algorithm is shared between contour and contourf.
>> For well-behaved datasets this is normally not a problem, but it becomes
>> obvious if you contour a random array.