Hey everyone, when plotting the cluster assignment/labels output from scikit’s k-means with matplotlib one can do this:

plt.scatter(sqrt(X[:,0]), sqrt(X[:,1]), s = 100, c = k_means.labels_ , alpha=0.4)

plt.autoscale(tight=True)

Matplotlib will chose colors from the k_means.labels array that might look like [0,1,2,0,1,0,0,0,2]

Doing something similar in Bokeh, doesn’t work:

from bokeh.plotting import *

output_notebook()

scatter(X[:,0],X[:,1], c = k_means.labels_ , alpha=0.4,

fill_alpha=0.6, name=“color_scatter_example”, title = “k-means clusters”)

Is there something better than manually mapping the color with an array that explicitly defines colors like:

color_list =

for i in k_means.labels_:

if i == 0:

color_list.append(’#d18096’)

elif i ==1:

color_list.append(’#483496’)

else:

color_list.append(’#00FFD0’)

Hey Kevin,

The bad news is that you do have to manually map the colors, but the good news is that it can be done relatively easily with Numpy. Below is a code sample.

-Peter

from bokeh.plotting import *

from numpy import *

N = 9

x = np.linspace(-2, 2, N)

y = x**2

output_file(“glyphs.html”, title=“glyphs.py example”)

color_list = array((‘#d18096’, ‘#483496’, ‘#00FFD0’))

labels = [0,1,2,0,1,0,0,0,2]

Materialize the full list of colors, using Numpy’s “fancy indexing”

(which is why we needed to create color_list as a numpy array)

colors = color_list[labels]

scatter(x, y, type=“circle_x”, size=16,

fill_color=colors, fill_alpha=0.6, alpha=0.4)

show()