Thread by @Josh_Ebner

If you want to create great data visualizations, you need to understand color palettes.

Here are a few quick tips:

[1/n]

#datascience #datavisualization #Python #rstats

[2/n]

For data that has a sequential ordering (i.e., low to high), you should use sequential color scales.

matplotlib.org/stable/tutoria…

#Python #matplotlib

[3/n]

Sequential color scales incrementally change saturation or lightness.

For example, this is a red-sequential color palette:

[4/n]

Having said that, there are a variety of sequential color palettes in Matplotlib (and a similar set in R).

#datascience #datavisualization #Python #rstats

[5/n]

There are also "perceptually uniform" sequential palettes.

In a perceptually uniform palette, equal steps in data are perceived as equal steps in color space.

#datascience #datavisualization

[6/n]

The end result, is that visualizations with perceptually uniform color palettes are typically much easier to interpret.

They're also typically very beautiful (I created these in R using the "viridis" palette).

#datascience #datavisualization #rstats

[7/n]

Now, let's talk about a different type of color palette: diverging palettes.

#data #datascience #DataVisualization

[8/n]

Diverging palettes typically have three main colors:

– 2 anchor colors (one at either end of the palette)
– 1 midpoint color (somewhere in the middle of the palette)

Between the anchor colors, the palette changes and the colors "meet" in the middle at a central color.

[9/n]

Here are a few examples of diverging color palettes

seaborn.pydata.org/tutorial/color…

#data #datascience #DataVisualization

[10/n]

Notice in the examples above that the two anchor colors are more "colorful"

And the central color is typically black, white, or grey.

[11/n]

In more technical terms, the two end colors are "saturated" colors.

And the midpoint color is a "desaturated" color.

[12/n]

That's typically how a diverging palette works:

The two end colors are "saturated" "colorful" hues (like red, blue, purple, etc) ...

[13/n]

And the palette changes decreases saturation until it reaches the midpoint, where there's typically a desaturated color like white, black, or grey.

#data #datascience #DataVisualization

[14/n]

There are also diverging palettes that change *lightness* instead of or along with saturation.

Common examples of this are diverging palettes with yellow at the midpoint.

#data #datascience #DataVisualization

[15/n]

So when do we use diverging palettes?

As you might guess, the best use of diverging palettes is when there is a natural midpoint.

#data #datascience #DataVisualization

[16/n]

So for example, you can use a diverging palette where you the data have a bad/OK/good ranking.

Here's one such visualization I made of the "Best States For Business":

[17/n]

Diverging palettes are perhaps most commonly used in US political maps, where we show % republican vs % democrat

Here's a beautiful example of an election map by @htmldon that uses a Red-Grey-Blue diverging palette.

[18/n]

Note:

You can find my full analysis of that visualization in this thread:

https://twitter.com/Josh_Ebner/status/1337500632895410179?s=20

[19/n]

Diverging color palettes can be a little harder to use, because you sometimes need to wrangle your data so the right colors map to the right data values.

[20/n]

This is particularly true for setting the midpoint color.

Sometimes the midpoint color is set to 0, and sometimes it's set to something else like a mean or median value.

[21/n]

So diverging color palettes can be a little harder to use.

[22/n]

But if you use them properly ...

Diverging palettes are extremely valuable for telling certain types of stories with your data.

[23/n]

Again: In particular, you often need diverging palettes for any data where there are natural endpoints, along with a natural midpoint.

#data #datascience #DataVisualization

Source, Matplotlib:

matplotlib.org/stable/tutoria…