quantile plot

plot_quantiles.py

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

def get_bins(scores, q=10, adjust_endpoints=False):
    _, bins =  pd.qcut(scores, q=q, retbins=True)
    if adjust_endpoints:
        bins[0] = -0.1
        bins[len(bins)-1] = 1.1
    return bins


def get_quantiles_from_bins(scores, bins, one_high=True):
    quantiles = pd.Series(np.digitize(scores, bins, right=True))
    if one_high:
        q = len(bins)-1
        quantiles = q - quantiles + 1
    return quantiles

def plot_quantiles(target, probs, q=10, figsize=(8,5),
                   title='Quantile Bar Graph'):

    if q==10:
        quant_type = 'Decile'
    elif q==20:
        quant_type = 'Ventile'
    else:
        quant_type = 'Quantile'

    # Get quantiles from probabilities
    bins = get_bins(probs, q=q, adjust_endpoints=True)
    quantiles = get_quantiles_from_bins(probs, bins, one_high=True)

    # Calculate event rate for each quantile
    quantile_labels = sorted(set(quantiles))
    rates = []
    for q in quantile_labels:
        arr = target[quantiles == q]
        pos_vals = arr.sum()
        rate = pos_vals / float(len(arr))
        rates.append(rate)

    # Plot event rate for each quantiles
    fig, ax = plt.subplots(figsize=figsize)
    ax.bar(quantile_labels, rates, align='center')
    ax.set_xticks(quantile_labels)
    plt.title(title)
    plt.ylabel('Event Rate')
    plt.xlabel(quant_type)
    plt.close(fig)
    return fig
    
    
# Keep in mind that this is set to create 10 deciles. If you want more than 10, change the parameter in the "get_bins" and "plot_quantiles" function
# Example of how to run this - 

plot_quantiles(y_test, scores)