On Fri, Apr 2, 2010 at 11:42 AM, Eric Firing <efir...@hawaii.edu> wrote: > Ryan May wrote: >> >> On Fri, Apr 2, 2010 at 1:23 AM, Eric Firing <efir...@hawaii.edu> wrote: >>>> >>>> On Fri, Mar 26, 2010 at 12:13 PM, Ryan May <rma...@gmail.com> wrote: >>>>> >>>>> I just hit a problem with using quiver with Basemap when when >>>>> angles='xy'. Because Basemap's x,y units are in meters, you end up >>>>> with angles that are quantized due to floating point truncation >>>>> (30000. + 0.001*u = 30000.). Changing to angles='uv' fixes the >>>>> problem, but it probably should be automatically scaled, as noted in >>>>> the comments: >>>>> >>>>> elif self.angles == 'xy' or self.scale_units == 'xy': >>>>> # We could refine this by calculating eps based on >>>>> # the magnitude of U, V relative to that of X, Y, >>>>> # to ensure we are always making small shifts in X, Y. >>>>> >>>>> I managed to fix the problem locally by setting: >>>>> >>>>> angles, lengths = self._angles_lengths(U, V, eps=0.0001 * >>>>> self.XY.max()) >>>>> >>> I don't think this will work in all cases. For example, there could be a >>> single arrow at (0,0). >> >> Good point. >> >>> Instead of self.XY.max(), how about abs(self.ax.dataLim.width)? >> >> Wouldn't this have problems if we zoom in sufficiently that the width >> is much less than magnitude of the values? Not exactly sure what data >> set would sensibly yield this, so I'm not sure if we should worry >> about it. >> >> If we do care, we could just put a minimum bound on eps: >> >> eps=max(1e-8, 0.0001 * self.XY.max()) > > I don't like taking the max of a potentially large array every time; and one > needs max absolute value in any case. I think the following is better: > > eps = np.abs(self.ax.dataLim.extents).max() * 0.001
I hadn't thought about performance. I think that's more important than any worries about bounds being disproportionately smaller. I'll check this in. Ryan -- Ryan May Graduate Research Assistant School of Meteorology University of Oklahoma ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
