Plot that depicts generalization problems that can occur when training a classifier.

generalization.py

import numpy
import pylab

steps = 100
zeros = numpy.zeros(steps)
ideal = numpy.linspace(0, 1, steps)

pylab.figure()
pylab.xlabel("Training Error")
pylab.ylabel("Test Error")
pylab.fill_between(ideal, zeros, ideal)
pylab.text(0.03, 0.02, "Theoretically impossible", verticalalignment="bottom",
           horizontalalignment="left", color="white", fontsize=10)
pylab.fill_between(ideal, ideal, ideal+0.1, color="g")
pylab.text(0.05, 0.15, "Optimum", verticalalignment="bottom", horizontalalignment="left",
           color="white", fontsize=10)
pylab.fill_between(ideal, ideal+0.1, 1, color="r")
pylab.text(0.02, 0.5, "Overfitting", verticalalignment="bottom", horizontalalignment="left",
           color="k", fontsize=10)
pylab.fill_between(numpy.array([0.2, 1]), 0, 1, color="orange")
pylab.text(0.8, 0.5, "Underfitting", verticalalignment="bottom", horizontalalignment="right",
           color="k", fontsize=15)
pylab.gca().set_xlim([0, 1])
pylab.gca().set_ylim([0, 1])
pylab.legend()
pylab.show()