separation_plot.py

# separation plot
# Author: Cameron Davidson-Pilon,2013
# see https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5907.2011.00525.x


import matplotlib.pyplot as plt
import numpy as np



def separation_plot( p, y, **kwargs ):
    """
    This function creates a separation plot for logistic and probit classification. 
    See https://onlinelibrary.wiley.com/doi/10.1111/j.1540-5907.2011.00525.x
    
    p: The proportions/probabilities, can be a nxM matrix which represents M models.
    y: the 0-1 response variables.
    
    """    
    assert p.shape[0] == y.shape[0], "p.shape[0] != y.shape[0]"
    n = p.shape[0]

    try:
        M = p.shape[1]
    except:
        p = p.reshape( n, 1 )
        M = p.shape[1]

    colors_bmh = np.array( ["#eeeeee", "#348ABD"] )


    fig = plt.figure( )
    
    for i in range(M):
        ax = fig.add_subplot(M, 1, i+1)
        ix = np.argsort( p[:,i] )
        #plot the different bars
        bars = ax.bar( np.arange(n), np.ones(n), width=1.,
                color = colors_bmh[ y[ix].astype(int) ], 
                edgecolor = 'none')
        ax.plot( np.arange(n+1), np.append(p[ix,i], p[ix,i][-1]), "k",
                 linewidth = 1.,drawstyle="steps-post" )
        #create expected value bar.
        ax.vlines( [(1-p[ix,i]).sum()], [0], [1] )
        plt.xlim( 0, n)
        
    plt.tight_layout()
    
    return