ad_pai_models/q_plot_tool.py

#! /usr/bin/env python
# -*- coding: utf-8 -*-
# vim:fenc=utf-8
#
# Copyright © 2025 StrayWarrior <i@straywarrior.com>

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

plt.rcParams['font.sans-serif'] = ['Songti SC']

def draw_figures(df, plot_name, sample_rate=1, filename=None):
    num_bins = 20
    df['p_bin'], _ = pd.qcut(df['score'], q=num_bins, duplicates='drop', retbins=True)
    quantile_data = df.groupby('p_bin').agg(
        mean_p=('score', 'mean'),
        mean_y=('label', 'mean')
    ).reset_index()
    
    ctr = quantile_data['mean_y']
    actual_quantiles = ctr / (ctr + (1 - ctr) / sample_rate)

    pctr = quantile_data['mean_p']
    predicted_quantiles = pctr / (pctr + (1 - pctr) / sample_rate)

    plt.figure(figsize=(6, 6))
    plt.plot(predicted_quantiles, actual_quantiles, ms=3, ls='-', color='blue', label='old')
    plt.plot([0, 1], [0, 1], color='gray', linestyle='--', label='Ideal Line')
    axis_max = max(predicted_quantiles.tolist()[-1],
                   actual_quantiles.tolist()[-1])
    plt.xlim(0, axis_max)
    plt.ylim(0, axis_max)
    plt.xlabel('Predicted pCTR')
    plt.ylabel('Actual CTR')
    plt.title('Q-Q Plot for pCTR Calibration %s' % (plot_name))
    plt.grid(True)
    if filename:
        plt.savefig(filename)
    else:
        plt.show()