How does Holoviews know which colours to assign to each scatterplot in an overlay?

Question

In the bokeh Holoviews gallery, there is an example called 'Scatter economic'.

http://holoviews.org/gallery/demos/bokeh/scatter_economic.html#bokeh-gallery-scatter-economic

In this plot, notice how one of the options for Scatter is ( color=Cycle('Category20') ). The last line of the plot is gdp_unem_scatter.overlay('Country') .

• My question is: How does Holoviews know to connect each Scatter to a particular color in Cycle('Category20') ? Is this just a property of Cycle() ? Is there some way that the Overlay interacts with the Scatter and with the Cycle automatically?
• A slightly related confusion is that if I use the .opts method instead of the cell magic as in the example, it still works. For example, if I use the .opts method with this cycle color on the Scatter (ie, second to the last line in the above example), and then do an .overlay('Country') , somehow Holoviews knows to assign each Scatter to a particular color based on the Country.

I want to make sure that I am properly plotting what I intend to.

Thank you!

Answer 1

How does Holoviews know to connect each Scatter to a particular color in Cycle('Category20')? Is this just a property of Cycle()? Is there some way that the Overlay interacts with the Scatter and with the Cycle automatically?

You are correct that Cycle and Overlay are designed to interact in this way automatically. More explicitly, each color in the Cycle gets assigned to a 'layer' of the overlay until the cycle runs out of colors and it loops.

For example, if I use the .opts method with this cycle color on the Scatter (ie, second to the last line in the above example), and then do an .overlay('Country'), somehow Holoviews knows to assign each Scatter to a particular color based on the Country.

This is because your call to opts customizes the options on the elements of the data structure before you call the overlay method on (this data structure is a HoloMap ). The options set there are propagated to the Scatter elements in the HoloMap which will now have the chosen Cycle specified. This means that when these elements get put into an overlay, HoloViews can look up the Cycle appropriately and apply it correctly to the overlay.

Hope that makes sense!

Answer 2

It is now possible to map categories in an NdOverlay (as is used in the example above) by using a so called dim expression and then define an expression to do the mapping:

dim_expr = hv.dim('category').categorize({'A': 'red', 'B': 'green', 'C': 'blue'})
overlay = hv.NdOverlay({chr(65+i): hv.Scatter(np.random.rand(10, 2)) for i in range(3)}, 'category')
overlay.opts(hv.opts.Scatter(color=dim_expr))

In this example we created a dim expression which points to the 'category' dimension and then maps each category ('A', 'B' and 'C') to a color ('red', 'green', 'blue'). We then just assign that to the color option.