模型无关的局部解释（LIME）技术原理解析及多领域应用实践

# 导入必要的库
 
 # 加载数据集
 # 使用来自openml的德国信用数据集
 credit = fetch_openml('credit-g', version=1, as_frame=True)
 X = credit.data
 y = credit.target
 
 # 将目标变量转换为二进制（好/坏到0/1）
 y = y.map({'good': 0, 'bad': 1})
 
 # 预处理：使用独热编码将分类特征转换为数值
 X = pd.get_dummies(X, drop_first=True)
 
 # 分割数据
 X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
 
 # 标准化数据
 scaler = StandardScaler()
 X_train = scaler.fit_transform(X_train)
 X_test = scaler.transform(X_test)
 
 # 训练随机森林分类器
 model = RandomForestClassifier(n_estimators=100, random_state=42)
 model.fit(X_train, y_train)
 
 # 初始化LIME
 explainer = lime.lime_tabular.LimeTabularExplainer(
     training_data=X_train,
     feature_names=X.columns,
     class_names=['Good', 'Bad'],
     mode='classification'
 )
 
 # 遍历测试集中的多个实例
 for i in range(3):  # 可根据需要调整解释的实例数量
     # 打印实际记录
     actual_record = X_test[i]
     print(f"Actual record for instance {i}:")
     print(pd.DataFrame(actual_record.reshape(1, -1), columns=X.columns))
 
     # 生成LIME解释
     exp = explainer.explain_instance(X_test[i], model.predict_proba, num_features=5)
 
     # 显示LIME解释
     exp.show_in_notebook(show_table=True)
     exp.as_pyplot_figure()
     plt.show()
 
     # 提取并打印解释详情
     explanation = exp.as_list()
     print(f"Explanation for instance {i}:")
     for feature, weight in explanation:
         print(f"{feature}: {weight:.2f}")
     print("\n")

# 导入必要的库
 
 # 加载数据集
 file_path = '/content/sample_data/Hotel_Reviews.csv'
 data = pd.read_csv(file_path)
 
 # 合并'Negative_Review'和'Positive_Review'列
 negative_reviews = data[['Negative_Review']].rename(columns={'Negative_Review': 'Review'})
 negative_reviews['Sentiment'] = 'negative'
 positive_reviews = data[['Positive_Review']].rename(columns={'Positive_Review': 'Review'})
 positive_reviews['Sentiment'] = 'positive'
 
 # 连接正面和负面评论
 reviews = pd.concat([negative_reviews, positive_reviews])
 reviews = reviews[reviews['Review'].str.strip() != '']  # 移除空评论
 
 # 将标签编码为二进制
 reviews['Sentiment'] = reviews['Sentiment'].map({'positive': 1, 'negative': 0})
 
 # 分割数据
 X_train, X_test, y_train, y_test = train_test_split(reviews['Review'], reviews['Sentiment'], test_size=0.2, random_state=42)
 
 # 向量化文本数据并移除停用词
 vectorizer = TfidfVectorizer(max_features=1000, stop_words='english')
 X_train_tfidf = vectorizer.fit_transform(X_train)
 X_test_tfidf = vectorizer.transform(X_test)
 
 # 训练逻辑回归模型
 model = LogisticRegression()
 model.fit(X_train_tfidf, y_train)
 
 # 初始化LIME
 explainer = lime.lime_text.LimeTextExplainer(class_names=['negative', 'positive'])
 
 # 定义预测函数
 def predict_proba(texts):
     texts_transformed = vectorizer.transform(texts)
     return model.predict_proba(texts_transformed)
 
 # 遍历测试集中的多个实例
 for i in range(5):  # 可根据需要调整解释的实例数量
     # 打印实际评论
     actual_review = X_test.iloc[i]
     print(f"Explanation for instance {i}:")
     print(actual_review)
     
     # 生成LIME解释
     exp = explainer.explain_instance(actual_review, predict_proba, num_features=6)
     
     # 显示LIME解释
     exp.show_in_notebook()
     exp.as_pyplot_figure()
     plt.show()
 
     # 提取并打印解释详情
     explanation = exp.as_list()
     print(f"Explanation for instance {i}:")
     for phrase, weight in explanation:
         print(f"{phrase}: {weight:.2f}")
     print("\n")

# 导入必要的库
 
 # 加载MNIST数据集（用作制造组件图像的代理）
 (X_train, y_train), (X_test, y_test) = mnist.load_data()
 
 # 预处理数据
 X_train = X_train.reshape(-1, 28, 28, 1).astype('float32') / 255
 X_test = X_test.reshape(-1, 28, 28, 1).astype('float32') / 255
 y_train = to_categorical(y_train, 10)
 y_test = to_categorical(y_test, 10)
 
 # 训练一个简单的CNN模型
 model = Sequential([
     Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(28, 28, 3)),
     MaxPooling2D(pool_size=(2, 2)),
     Flatten(),
     Dense(128, activation='relu'),
     Dense(10, activation='softmax')
 ])
 model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
 model.fit(X_train_rgb, y_train, epochs=5, batch_size=200, verbose=1, validation_data=(X_test_rgb, y_test))
 
 # 初始化LIME
 explainer = lime.lime_image.LimeImageExplainer()
 
 # 定义预测函数
 def predict_proba(images):
     return model.predict(images)
 
 # 选择实例进行解释
 for i in range(5):  # 可根据需要调整解释的实例数量
     # 获取一个实例进行解释
     image = X_test_rgb[i]
     explanation = explainer.explain_instance(image, predict_proba, top_labels=1, hide_color=0, num_samples=1000)
     
     # 获取顶级标签的解释
     temp, mask = explanation.get_image_and_mask(explanation.top_labels[0], positive_only=True, num_features=5, hide_rest=False)
     
     # 显示带有解释的图像
     plt.imshow(mark_boundaries(temp, mask))
     plt.title(f"Explanation for instance {i}")
     plt.show()
 
     # 打印详细解释
     print(f"Explanation for instance {i}:")
     print(explanation.local_exp[explanation.top_labels[0]])
     print("\n")