Post modeling: create probability calibration curves

[rayidghani]

http://scikit-learn.org/stable/auto_examples/calibration/plot_calibration_curve.html

[shaycrk]

From #654:

Lower priority, but one plot I find helpful and easy to explain to non-technical audiences is the stack-ranking performance of the model (e.g., splitting the test set by deciles or vigintiles and looking at how the average score compares to the average label in each as a bar graph). This can be used to look at calibration or simply show that the model is ranking properly and how the propensity differs from top to bottom.