📖 第18课：分类模型评估

from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.linear_model import LogisticRegression from sklearn.model_selection import cross_val_predict from sklearn.metrics import (classification_report, confusion_matrix, precision_recall_fscore_support, accuracy_score) import numpy as np texts = [ "这部电影非常好看剧情精彩", "太棒了强烈推荐", "令人感动的故事", "音乐优美画面精致", "经典之作百看不厌", "温馨治愈心情愉悦", "这部电影太差了浪费时间", "剧情无聊看不下去", "演技尴尬令人失望", "太烂了简直受不了", "无聊至极毫无亮点", "烂片无疑别去看", "还行吧一般般", "有点无聊但能看", "中规中矩没什么惊喜", ] labels = ['pos'] * 6 + ['neg'] * 6 + ['neu'] * 3 vectorizer = TfidfVectorizer(token_pattern=r'(?u)\b\w+\b') X = vectorizer.fit_transform(texts) clf = LogisticRegression(max_iter=1000, random_state=42) y_pred = cross_val_predict(clf, X, labels, cv=3) print("=== 分类评估报告 ===") print(classification_report(labels, y_pred)) # 混淆矩阵 cm = confusion_matrix(labels, y_pred, labels=['pos', 'neg', 'neu']) print("混淆矩阵:") print(f"{'':8s} pred_pos pred_neg pred_neu") for i, true_label in enumerate(['pos', 'neg', 'neu']): print(f"{true_label:8s} {cm[i][0]:8d} {cm[i][1]:8d} {cm[i][2]:8d}") # 逐指标分析 p, r, f1, s = precision_recall_fscore_support(labels, y_pred, average=None, labels=['pos', 'neg', 'neu']) print("\n=== 逐类别指标 ===") for i, label in enumerate(['pos', 'neg', 'neu']): print(f" {label}: P={p[i]:.3f} R={r[i]:.3f} F1={f1[i]:.3f} support={s[i]}") # macro vs micro p_macro, r_macro, f1_macro, _ = precision_recall_fscore_support(labels, y_pred, average='macro') p_micro, r_micro, f1_micro, _ = precision_recall_fscore_support(labels, y_pred, average='micro') print(f"\nmacro: P={p_macro:.3f} R={r_macro:.3f} F1={f1_macro:.3f}") print(f"micro: P={p_micro:.3f} R={r_micro:.3f} F1={f1_micro:.3f}")

import numpy as np from sklearn.metrics import roc_curve, auc from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.linear_model import LogisticRegression from sklearn.model_selection import train_test_split # 二分类ROC texts = [ "这部电影好看剧情精彩", "太棒了推荐", "感动故事", "优美精致经典", "满意治愈愉悦", "好看得停不下来", "太差了浪费时间", "无聊看不下去", "尴尬失望", "烂片受不了", "廉价混乱", "后悔别去", ] labels = [1]*6 + [0]*6 # 1=pos, 0=neg vectorizer = TfidfVectorizer(token_pattern=r'(?u)\b\w+\b') X = vectorizer.fit_transform(texts) X_train, X_test, y_train, y_test = train_test_split(X, labels, test_size=0.3, random_state=42) clf = LogisticRegression(max_iter=1000, random_state=42) clf.fit(X_train, y_train) y_prob = clf.predict_proba(X_test)[:, 1] fpr, tpr, thresholds = roc_curve(y_test, y_prob) roc_auc = auc(fpr, tpr) print("=== ROC曲线数据 ===") print(f"AUC = {roc_auc:.4f}") print(f"{'阈值':8s} {'FPR':8s} {'TPR':8s}") for t, f, tp in zip(thresholds, fpr, tpr): print(f"{t:8.4f} {f:8.4f} {tp:8.4f}") # 精确率-召回率权衡 print("\n=== 精确率-召回率权衡 ===") for threshold in [0.3, 0.5, 0.7]: y_pred_t = (y_prob >= threshold).astype(int) tp = sum((y_pred_t == 1) & (np.array(y_test) == 1)) fp = sum((y_pred_t == 1) & (np.array(y_test) == 0)) fn = sum((y_pred_t == 0) & (np.array(y_test) == 1)) prec = tp / (tp + fp) if (tp + fp) > 0 else 0 rec = tp / (tp + fn) if (tp + fn) > 0 else 0 print(f" 阈值={threshold:.1f}: P={prec:.3f} R={rec:.3f} F1={2*prec*rec/(prec+rec) if (prec+rec) > 0 else 0:.3f}")