prediction - Measuring "mixtureness" of labeled data (python) -

i have 2d data:

the data labeled , shown in different colors. non supervised process not yield correct prediction because data pretty mixed (although colors seem have regions of preference). want see if possible measure how mixed points different sets.

for need define measurement of how mixed (i think should exist). nice have these algorithms implemented. looking simple predictive model can trained used data shown. help. if possible i'm looking these implementations in python.