How to shade an image with quantile color scale?

I’m trying to plot points with DataShader and want to render points with a quantile color scale. Would you please help me with it?

Data:

lon,lat,metric
30.771635, 108.353681, 34
30.765280, 108.351067, 85 

Code:

import datashader as ds
import datashader.transfer_functions as tf
import colormap_select from datashader.colors 

// shade image
cm = partial(colormap_select)
agg = cvs.points(df, lon, lat, ds.sum('metric'))
img = tf.shade(agg, cmap=cm(['green', 'yellow', 'red']), how='eq_hist')

My Goal [‘green’, ‘yellow’, ‘red’] represent a quantile scale of aggregation: [0, 33%, 66%, 100%] . In other words:

• green color: min -> 33% * max
• yellow color: 33% * max -> 66% * max
• red color: 66% * max -> max

Is there a method to achieve this color scale? furthermore, is there a way to flexibly define color stops ([0, 33%, 66%, 100%]) ?

Thank you.

[jbednar]

Your code above can be simplified to:

import datashader as ds
import datashader.transfer_functions as tf

agg = cvs.points(df, lon, lat, ds.sum('metric'))
img = tf.shade(agg, cmap=['green', 'yellow', 'red'], how='eq_hist')

And presumably you are asking to get rid of the interpolation between green, yellow, and red that your code will produce?

If so, you can do that by constructing a colormap that avoids interpolation:

But personally, I like the interpolated one better. 😃

[jbednar]

Glad you like our work! The example above is an adaptation and extension of my “census_gray_redhot1_eq_hist” example in http://datashader.org/topics/census.html. But that’s just one example buried deep in between a bunch of others, and I think colormap manipulations like that are worth addressing explicitly in the user manual so that more people realize what can be done.

We’re currently working on re-implementing datashader’s more advanced colormap features in Bokeh, and when we have that ready, we should include an overview in the manual of how to use colormapping to reveal your data flexibly and effectively, particularly in conjunction with the percentile interpretation of the results of eq_hist.