ROC、PR曲线绘制

PR曲线绘制

https://blog.csdn.net/xiang_gina/article/details/149962713 

三分钟带你理解ROC曲线和PR曲线

https://blog.csdn.net/Guo_Python/article/details/105820358

https://developers.google.com/machine-learning/crash-course/overfitting/interpreting-loss-curves?hl=zh-cn

---

 Cursor编写的示例程序：

提示：写一个Demo，计算ROC、AUC、和PR曲线，要求，计算过程中通过阈值按照步长逐渐增长的方式（步长采用0.01）

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import numpy as np
import matplotlib.pyplot as plt


def confusion_from_threshold(y_true, y_scores, threshold):
    """Compute TP/FP/TN/FN for rule: predict positive if score >= threshold."""
    y_true = np.asarray(y_true).astype(int)
    y_scores = np.asarray(y_scores).astype(float)

    y_pred = (y_scores >= threshold).astype(int)

    tp = int(np.sum((y_pred == 1) & (y_true == 1)))
    fp = int(np.sum((y_pred == 1) & (y_true == 0)))
    tn = int(np.sum((y_pred == 0) & (y_true == 0)))
    fn = int(np.sum((y_pred == 0) & (y_true == 1)))

    return tp, fp, tn, fn


def sweep_thresholds(y_true, y_scores, step=0.1):
    """Sweep threshold from 0.0 to 1.0 (inclusive) with given step."""
    thresholds = np.round(np.arange(0.0, 1.0 + 1e-9, step), 10)

    total_p = int(np.sum(y_true))
    total_n = int(len(y_true) - total_p)

    rows = []
    for thr in thresholds:
        tp, fp, tn, fn = confusion_from_threshold(y_true, y_scores, thr)

        tpr = tp / total_p if total_p > 0 else 0.0
        fpr = fp / total_n if total_n > 0 else 0.0

        precision = tp / (tp + fp) if (tp + fp) > 0 else 1.0
        recall = tpr

        rows.append(
            {
                'thr': float(thr),
                'tp': tp,
                'fp': fp,
                'tn': tn,
                'fn': fn,
                'tpr': float(tpr),
                'fpr': float(fpr),
                'precision': float(precision),
                'recall': float(recall),
            }
        )

    return rows


def trapezoid_auc(x, y):
    """Area under curve using trapezoidal rule after sorting by x."""
    x = np.asarray(x, dtype=float)
    y = np.asarray(y, dtype=float)

    order = np.argsort(x)
    x = x[order]
    y = y[order]

    return float(np.trapz(y, x))


def main():
    # Built-in example (10 samples)
    scores = np.array([0.95, 0.85, 0.78, 0.66, 0.60, 0.55, 0.53, 0.52, 0.51, 0.40], dtype=float)
    y_true = np.array([1, 1, 0, 1, 0, 1, 0, 0, 1, 0], dtype=int)  # P=1, N=0

    step = 0.01
    rows = sweep_thresholds(y_true, scores, step=step)

    # Collect curve points
    fprs = [r['fpr'] for r in rows]
    tprs = [r['tpr'] for r in rows]
    recalls = [r['recall'] for r in rows]
    precisions = [r['precision'] for r in rows]

    roc_auc = trapezoid_auc(fprs, tprs)

    # Print table
    print(f"Threshold sweep step={step}")
    print("thr\tTP\tFP\tTN\tFN\tTPR\tFPR\tPrec\tRecall")
    for r in rows:
        print(
            f"{r['thr']:.1f}\t{r['tp']}\t{r['fp']}\t{r['tn']}\t{r['fn']}\t"
            f"{r['tpr']:.3f}\t{r['fpr']:.3f}\t{r['precision']:.3f}\t{r['recall']:.3f}"
        )

    print('\nROC AUC (threshold-sweep, trapezoid) =', roc_auc)

    # Plot
    fig, axes = plt.subplots(1, 2, figsize=(12, 5))

    axes[0].plot(fprs, tprs, marker='o', label=f'ROC (AUC={roc_auc:.4f})', color='tab:blue')
    axes[0].plot([0, 1], [0, 1], linestyle='--', linewidth=1, color='gray', label='Random')
    axes[0].set_xlabel('False Positive Rate')
    axes[0].set_ylabel('True Positive Rate')
    axes[0].set_title('ROC Curve (threshold sweep)')
    axes[0].set_xlim(0, 1)
    axes[0].set_ylim(0, 1)
    axes[0].grid(True)
    axes[0].legend(loc='lower right')

    axes[1].plot(recalls, precisions, marker='o', label='PR (threshold sweep)', color='tab:red')
    axes[1].set_xlabel('Recall')
    axes[1].set_ylabel('Precision')
    axes[1].set_title('PR Curve (threshold sweep)')
    axes[1].set_xlim(0, 1)
    axes[1].set_ylim(0, 1)
    axes[1].grid(True)
    axes[1].legend(loc='lower left')

    plt.tight_layout()
    plt.show()


if __name__ == '__main__':
    main()

结果：