Updated the code to include support for matplotlib colormaps

[pakitochus]

By default it uses colormap='rainbow' to generate colors, so that there is no error if len(x)>10. A list of color can still be specified, and if len(x) not > 10 it will apply. hex2rgb has changed to numpy array to comply with matplotlib colormaps specification.

[fengwangPhysics]

@pakitochus Thanks for your interest in the code for chord diagram. I have a few comments on your changes, please check and update accordingly. Thanks.

[fengwangPhysics]

This "else" will apply only when len(x)<=10 and (colors is not None). The colors will be changed in the following codes, hence this seems to be a bug.

My suggestion:

if colors is None:
    if (len(x) > 10) or (cmap is not None):
        cm = plt.get_cmap(cmap)
        colors = cm(np.linspace(0,1,len(x)))[:,:3]
    else:
        # use default d3.js category10

[pakitochus]

In this case, it is when if (len(x) > 10) or (colors is None):, implying or, not and. I mean, if any of these two conditions are met, we use the default colormap. I don't know if this is what you meant. Your suggestion would have issues when len(x)>10 and len(colors)<=10, although we could work out a solution.

[fengwangPhysics]

In my original version, I assume that users make sure that len(colors)=len(x) , if they use colors. Therefore, I didn't expect the issue len(x)>10 and len(colors)<=10.

My original idea is that:
(1) if colors is None
(1a) if len(x)<=10, use d3.js category10 colors;
(1b) else, quit. I used to think that it is not a good idea to plot a chord diagram with too many categories, which makes the chords too many to see individual ones. Maybe I am wrong at this point. At lease, I have the keyword colors to rescue this.
(2) if colors is given by users, len(colors)=len(x), and the colors are sometimes meaningful, e.g., colors are chosen to be consistent with other panels. In this case, everything is left to the users.

We can think about the following options:

1. Maybe we can use your cmap if len(x)>10, and d3.js category10 if len(x)<10?
2. Maybe we can use d3.js category20 colors if 10<len(x)<=20; and raise an error if len(x)>20?

Let's also assert len(colors)==len(x) if not colors is None.

What do you think?

[pakitochus]

Yes, I think the best idea is prioritize "colors" vs "colormap". I mean, by default (with no colormap or colors specified) use category10 if len(x)<=10, category20 if 10<len(x)<=20 and gist_rainbow (better than rainbow) if len(x)>20. It is indeed difficult to visualize if there are more than 15 categories, but sometimes it is inevitable.

I guess I will remove the colormap option and use colors instead. I mean:

if type(colors) is list:
      # use the list of colors (and `assert len(colors)==len(x)`)
elif type(colors) is str:
      # use the colormap specified
else: 
      #use the default colors specified

It should be easier that way. What do you think about this solution?

[fengwangPhysics]

Great idea. I think this solution will also simplify the user interface for the function. Please implement it. Thanks a lot.

[fengwangPhysics]

plt has already been imported in line 3.

[pakitochus]

Sure! I'll change that.

[fengwangPhysics]

The d3.js category10 colors have been used as the default colors for matplotlib v2.x. It has a better visualization than sequential color map to represent category data.

[pakitochus]

Ok, is there any colormap in matplotlib including the d3.js category10? I couldn't find them..

[fengwangPhysics]

There is no color map for d3.js category10 in matplotlib. They are used as the default color cycles for plt.plot(). Please check this: https://matplotlib.org/users/dflt_style_changes.html#colors-color-cycles-and-color-maps

Usually colormap is used for continuous variables, rather than category data. Therefore, my opinion is to use category10 to represent the data if there are less than 10 categories.

[pakitochus]

Perfect then, we'll stick with category10 if less than 10 categories and no colormap specified.

[pakitochus]

The requested changes have been added. I slightly changed the code, prioritizing category10 and asserting afterwards if len(x)<=len(colors) to raise an error if there were more categories than colors. Added option for colormap as keyword in the same variable (colors).

[fengwangPhysics]

Thanks. The changes in matplotlib-chord.py is good. Sorry for the late reply.

I don't have experience on PR, so I am not sure whether it will interfere your PR if I directly change README.md. So please also update README.md and example.png accordingly.