Allow more customization in GraphPlot

[ppurka]

This is my suggestion of problems and possible enhancements of graph plotting:

• For now we say either vertex_colors='red' or vertex_colors={'red': [1], 'blue': [2, 3]}. But contrary to this we have edge_color and edge_colors. Not good.

• Default vertex color is pink. Setting color of some vertices will change color for other vertices to light blue. Not good.

• fontsize affects to size of title font, not for vertex label or edge label font sizes.

• We have no way at all to change, say, fontsize of vertex labels or color of edge labels etc.

• And why we can have different colors in vertices, but not shapes etc?

Suggestions

• Make pink true default color for vertices, so that setting color of vertex 1 will left vertex 2 to pink. (Done in #21048)

• Add option vertex_color that will be default color for vertices not colored with vertex_colors. Deprecate use of vertex_colors='red'. (Done in #21048)

• Allow using edge_color and edge_colors at same time like edge_color='red', edge_colors={'blue': [(1, 2)]}. (For now setting edge_colors will change color of other edges to black, overriding edge_color.) (Done in #21053)

• Add title_fontsize. Deprecate fontsize.

• Add other options with same pattern. I.e. let we still have vertex_shape, but also add vertex_shapes to be used like G.plot(vertex_shape='s', vertex_shapes={'d': G.sources(), 'c': G.sinks()}).

• And last, we should decide how to give options for fonts. As a triples of color, font and fontsize, or as a three argument? If as triples, do not add title_fontsize but title_font.

First action point would be to think about my suggestion and modify it until we have common vision of what to do.

Above is a copy from dicussion, to be found easily. Here is original description:

There is no customization of text possible in plots of graphs (graphs as in graph theory). It is mentioned as a TODO in sage.graphs.graph_plot.GraphPlot.set_vertices

            # TODO: allow text options
            for v in self._nodelist:
                self._plot_components['vertex_labels'].append(text(str(v),
                    self._pos[v], rgbcolor=(0,0,0), zorder=8))

Here are some other bugs:

g = DiGraph({0:[1], 1:[0]})
g.show()  # OK
g.show(edge_colors={'red': [(0,1)]})  # Not OK
g.show(edge_style='dotted')  # OK

# Docstring says edge_style "currently only works for
# directed graphs" but the following works still.
graphs.CompleteGraph(3).show(edge_style="dotted")

# Even plain show() fails in extreme case:
g1 = graphs.CompleteGraph(2)
g2 = Graph({0:[1]})
print(g1==g2)  # True
g1.show()  # Not shown
g2.show()  # Shown

"For all the constructors in this database (except the octahedral, dodecahedral, random and empty graphs), the position dictionary is filled in" -- it is filled in octahedral and dodecahedral graph.

[jmantysalo]

• For now we say either vertex_colors='red' or vertex_colors={'red': [1], 'blue': [2, 3]}. But contrary to this we have edge_color and edge_colors. Not good.

• Default vertex color is pink. Setting color of some vertices will change color for other vertices to light blue. Not good.

• fontsize affects to size of title font, not for vertex label or edge label font sizes.

• We have no way at all to change, say, fontsize of vertex labels or color of edge labels etc.

• And why we can have different colors in vertices, but not shapes etc?

Suggestions

• Make pink true default color for vertices, so that setting color of vertex 1 will left vertex 2 to pink.

• Add option vertex_color that will be default color for vertices not colored with vertex_colors. Deprecate use of vertex_colors='red'.

• Allow using edge_color and edge_colors at same time like edge_color='red', edge_colors={'blue': [(1, 2)]}. (For now setting edge_colors will change color of other edges to black, overriding edge_color.)

• Add title_fontsize. Deprecate fontsize.

• Add other options with same pattern. I.e. let we still have vertex_shape, but also add vertex_shapes to be used like G.plot(vertex_shape='s', vertex_shapes={'d': G.sources(), 'c': G.sinks()}).

• And last, we should decide how to give options for fonts. As a triples of color, font and fontsize, or as a three argument? If as triples, do not add title_fontsize but title_font.

First action point would be to think about my suggestion and modify it until we have common vision of what to do.

[jmantysalo]

Just pinging... Any comments on my suggestions?

[ppurka]

Make pink true default color for vertices, so that setting color of vertex 1 will left vertex 2 to pink.

Maybe it should be blue/sage color, so as to be default with the sage plots.

Instead of adding many keyword arguments for the overall plot command, it might be better to let the methods take care of the setting.

• set_vertices could be used to set only vertex properties.
• set_edges for setting only edge property.
• Maybe introduce a different set_plot_font because it seems customizing all the different font properties is going to be a complicated method in itself.
• Introduce the vertex_shapes and such other vertex related properties not in plot but in the set_vertex method.

Adding kcrisman for his expert comments :-)

[jmantysalo]

Replying to ppurka:

Make pink true default color for vertices, so that setting color of vertex 1 will left vertex 2 to pink.

Maybe it should be blue/sage color, so as to be default with the sage plots.

Whatever. But I suppose that we agree that setting color of vertex 1 should not change color of vertex 2.

Instead of adding many keyword arguments for the overall plot command, it might be better to let the methods take care of the setting. - -

• Introduce the vertex_shapes and such other vertex related properties not in plot but in the set_vertex method.

so we would have something like

p = G.plot()
p.set_vertex(('shapes', {'d': G.sources()}), ('colors', ...))
p.show()

? Sounds complicated.

[ppurka]

I was thinking it would more like

p.set_vertex(shapes=..., colors=...)
p.set_edges(colors=...)
p.set_plot(fontsize=..., font=...)  # or set_plot_font, if the only extra thing left to customize is fonts

[jmantysalo]

Replying to ppurka:

I was thinking it would more like

p.set_vertex(shapes=..., colors=...)

But isn't it still much more complicated than just giving options in plot?

[ppurka]

In my opinion, it will get unwieldy if all the options are put into plot. Making a plot will involve a massive single line of p.plot(shapes=, colors=, fonts=,....). Breaking it up into multiple methods should make it clearer which methods control which aspects of the graph.

From the programming point of view, it should also make the functions more manageable. Otherwise, all of that code will end up in a single huge function plot.

[jmantysalo]

Replying to ppurka:

In my opinion, it will get unwieldy if all the options are put into plot. Making a plot will involve a massive single line of p.plot(shapes=, colors=, fonts=,....). Breaking it up into multiple methods should make it clearer which methods control which aspects of the graph.

It depends. User may not use all possible options. And if he/she uses, then it won't be that much easier with several lines of settings. And one can say

edge_colors = ... # something complicated
 ...
vertex_shapes = ... # something complicated
...plot(edge_colors=edge_colors, ...)

From the programming point of view, it should also make the functions more manageable. Otherwise, all of that code will end up in a single huge function plot.

I guess that backend should have set_edges and so on. But plot can call them.

I believe that there is not very good solution. There are just too many options that could be useful to someone. OTOH reducing possible options sounds odd. So we can only make sure that the design is as orthogonal as it can.

[jmantysalo]

Still about the desing... I the user is plotting just one graph, it is easier to have single plot() with massive list of options.

If the user wants to plot same graph somehow modified, then it will be easier to do something like

G.set_vertex_shape('a', 'rectangle')
G.set_vertex_shape('b', 'circle')
G.show()
G.do_something_inplace('a', 'x')
G.show()

From these choised I would choose first one. I would make a plot() with huge list of options, and internally set_vertices() and set_edges() with a manageable number of options.

[ppurka]

We can have both, can't we?

Make plot() be able to take multiple arguments and set them using the other custom functions. This will take care of my concerns in comment:13

[jmantysalo]

Replying to ppurka:

We can have both, can't we?

Make plot() be able to take multiple arguments and set them using the other custom functions. This will take care of my concerns in comment:13

Yes, that can be done.

[jmantysalo]

First part is at #21048.