Re: Comparing plots to Matlab

[Ben Abbott]

On Dec 1, 2010, at 5:42 PM, Søren Hauberg wrote:

> Hi All
> 
> For determining if our plotting system works, we currently have
> 'rundemos' which can be used to run the demos of our plotting functions.
> The results can then be visually inspected to see if they are good.
> However, I find it rather to figure out if the visual results are
> correct.
> 
> To somewhat ease this I made a script that behaves quite similarly to
> 'rundemos'. However, instead of running the demoes it generates a script
> that contains the demos along with code for printing the generated
> figures to PNG files. The current version of this script is attached.
> 
> I have then tried to run the generated script in Octave with the gnuplot
> backend (it crashes the FLTK backend) and in Matlab. This gives us a set
> of directly comparable PNG files. I have put these on
> 
>  http://octave.sourceforge.net/compare_plots/
> 
> In general, results seems quite good, though not perfect. On this page,
> you'll sometimes see missing figures. In the script I put try-catch'es
> around the individual demos, so if a demo fails no figure is generated.
> 
> Søren
> <dump_demos.m>

Nicely done!

Octave's colorbars look rather good in comparison to ML.

Some of the demos that set the data aspect ratio produce differences in the 
size of the axis box. However, all the data is present and the aspect ratio is 
correct, so I think this is ok.

Some demos that indicate problems are ...

        contour3_1_1.png

        errorbar_2_1.png

        hold_2_1.png

        hold_3_1.png

        hold_4_1.png

        stem_5_1.png

        stem_7_1.png

Perhaps some bug reports should be created once the bug tracker is back up.

Any reason why the plots are not ordered alphabetically?

Ben