Draw mean-difference plot

README.md

Explain

Mean-Difference plot, or Bland–Altman plot can be used to compare the prediction accuracy.

Bland and Altman drive the point that any two methods that are designed to measure the same parameter (or property) should have good correlation when a set of samples are chosen such that the property to be determined varies considerably.

This is how they looks:

The code is based on statsmodels.api but modified to change appereance. Also refer to: matplotlib - Bland-Altman plot in Python - Stack Overflow

draw.py

import pandas as pd
import statsmodels.api as sm
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import rcParams
from matplotlib import font_manager


def plot(m1, m2):
    # data process
    means = np.mean([m1, m2], axis=0)
    diffs = m1 - m2
    mean_diff = np.mean(diffs)
    std_diff = np.std(diffs, axis=0)

    # draw
    fig, ax = plt.subplots(1,figsize=(2.95,2.95), dpi=600) # 75mm/25.4=2.95磅
    ax.spines['right'].set_visible(False)
    ax.spines['top'].set_visible(False)

    ax.scatter(means, diffs, c='black', s=0.6) # Plot the means against the diffs.

    ax.axhline(mean_diff, color='gray', linewidth=1, linestyle='--')  # draw mean line.
    # Annotate mean line with mean difference.
    ax.annotate('mean diff:\n{}'.format(np.round(mean_diff, 2)),
                xy=(0.99, 0.5),
                horizontalalignment='right',
                verticalalignment='center',
                fontsize=7.5, # !!!
                xycoords='axes fraction')

    sd_limit=1.96
    half_ylim = (1.5 * sd_limit) * std_diff
    ax.set_ylim(mean_diff - half_ylim, mean_diff + half_ylim) # 对称
    limit_of_agreement = sd_limit * std_diff
    lower = mean_diff - limit_of_agreement
    upper = mean_diff + limit_of_agreement
    for j, lim in enumerate([lower, upper]):
        ax.axhline(lim, color='gray', linewidth=1, linestyle=':')
    ax.annotate(f'-{sd_limit} SD: {lower:0.2g}',
                xy=(0.99, 0.07),
                horizontalalignment='right',
                verticalalignment='bottom',
                fontsize=7.5, # !!!
                xycoords='axes fraction')
    ax.annotate(f'+{sd_limit} SD: {upper:0.2g}',
                xy=(0.99, 0.92),
                horizontalalignment='right',
                fontsize=7.5, # !!!
                xycoords='axes fraction')



    ax.set_xlabel('Mean of predicted and measured VO2\n($L×min^{-1}×kg^{-1}$)',fontsize=7.5)
    ax.set_ylabel('Difference of predicted and measured VO2\n($L×min^{-1}×kg^{-1}$)',fontsize=7.5)
    ax.tick_params(labelsize=7.5)
    fig.tight_layout()
    return ax

# data load
data = pd.read_csv("/content/Nanjing01_Plain.csv")
m1, m2 = data['y_true']/1000, data['y_pred']/1000
ax = plot(m1, m2)