import pandas as pd # 定义标背景框色函数 def background_color(val): color = 'red' if val < 60 else 'yellow' if val < 80 else 'green' return { 'background-color': color, 'border': '1px solid black' } # 创建DataFrame df = pd.DataFrame({ 'name': ['Alice', 'Bob', 'Charlie', 'David'], 'score': [70, 55, 85, 90] }) # 创建Styler对象并应用标背景框色函数 styler = df.style.applymap(background_color, subset=['score']) styler too many values to unpack (expected 2)

在你的代码中，出现了相同的错误。这是因为在计算aqi时，你尝试将Series对象（iaqi1和iaqi2）与标量值进行比较，导致了"The truth value of a Series is ambiguous"错误。 要解决这个问题，你需要使用逐元素的比较运算符来比较Series对象中的每个元素。在这种情况下，你可以使用Series对象的.max()方法来找到iaqi1和iaqi2中的最大值。 请修改以下代码： ```python aqi = max(iaqi1, iaqi2) ``` 为： ```python aqi = pd.Series([iaqi1, iaqi2]).max() ``` 这样就可以避免出现"The truth value of a Series is ambiguous"错误，并找到iaqi1和iaqi2中的最大值作为aqi。请确保在使用该代码之前导入pandas库（import pandas as pd）以使用pd.Series()方法。

好的！下面是基于您提供的代码，在另一个文件中实现 **Deep Domain Confusion (DDC)** 的完整解决方案，并将数据集划分部分与 DDC 实现分离到两个独立的 Python 文件中。 --- ### 第一步：`data_preparation.py` 这个文件负责加载和分割数据集： ```python # data_preparation.py import os import pandas as pd import numpy as np from sklearn.model_selection import train_test_split def process_target(): main_folder = 'C:/Users/Lenovo/Desktop/crcw不同端12k在0负载下/风扇端' data_list = [] label_list = [] for folder_name in sorted(os.listdir(main_folder)): folder_path = os.path.join(main_folder, folder_name) if os.path.isdir(folder_path): csv_files = [f for f in os.listdir(folder_path) if f.endswith('.csv')] print(f"Processing folder: {folder_name}, found {len(csv_files)} CSV files.") for filename in sorted(csv_files): file_path = os.path.join(folder_path, filename) csv_data = pd.read_csv(file_path, header=None) if csv_data.shape[1] >= 4: csv_data = csv_data.iloc[:, [0, 1, 2]].values else: print(f"Skipping file {filename}, unexpected shape: {csv_data.shape}") continue data_list.append(csv_data) if '内圈故障' in folder_name: class_label = 0 elif '球故障' in folder_name: class_label = 1 else: continue label_list.append(class_label) if data_list and label_list: data = np.array(data_list) # Shape: (num_samples, seq_length, num_features) labels = np.array(label_list) # Shape: (num_samples,) return data, labels else: raise ValueError("No valid data available to process.") def split_datasets(X, y, test_size=0.2, val_size=0.25): """ 数据集划分函数 参数: - X: 特征数据数组 - y: 标签数组 - test_size: 测试集比例，默认为 0.2 - val_size: 验证集占剩余训练数据的比例，默认为 0.25 """ X_train_val, X_test, y_train_val, y_test = train_test_split( X, y, test_size=test_size, stratify=y, random_state=42 ) X_train, X_val, y_train, y_val = train_test_split( X_train_val, y_train_val, test_size=val_size, stratify=y_train_val, random_state=42 ) return X_train, X_val, X_test, y_train, y_val, y_test if __name__ == "__main__": try: data0, label0 = process_target() # 划分数据集 X_train, X_val, X_test, y_train, y_val, y_test = split_datasets(data0, label0) print("Training Set:", X_train.shape, y_train.shape) print("Validation Set:", X_val.shape, y_val.shape) print("Testing Set:", X_test.shape, y_test.shape) # 将数据保存至 .npz 文件供其他脚本使用 np.savez('datasets.npz', X_train=X_train, y_train=y_train, X_val=X_val, y_val=y_val, X_test=X_test, y_test=y_test) except ValueError as e: print(e) ``` --- ### 第二步：`deep_domain_confusion.py` 此文件实现了 Deep Domain Confusion (DDC)，并从 `data_preparation.py` 中生成的数据集中读取内容： ```python # deep_domain_confusion.py import numpy as np import tensorflow as tf from tensorflow.keras.models import Model from tensorflow.keras.layers import Dense, Input, Lambda from tensorflow.keras.optimizers import Adam class DDCLoss(tf.keras.losses.Loss): """自定义损失函数用于计算领域混淆""" def call(self, domain_labels, domain_predictions): mmd_loss = compute_mmd(domain_labels, domain_predictions) return mmd_loss def compute_mmd(source_features, target_features): """ 最大均值差异（MMD）计算公式 MMD^2(P, Q) = || μ_P - μ_Q ||^2 """ source_mean = tf.reduce_mean(source_features, axis=0) target_mean = tf.reduce_mean(target_features, axis=0) return tf.reduce_sum(tf.square(source_mean - target_mean)) def build_model(input_shape): inputs = Input(shape=input_shape) # 共享特征提取层 shared_layer = Dense(64, activation='relu')(inputs) # 分类分支 classification_output = Dense(2, activation='softmax', name="classification")(shared_layer) # 域适应分支 domain_output = Dense(1, activation='sigmoid', name="domain")(shared_layer) model = Model(inputs=inputs, outputs=[classification_output, domain_output]) return model if __name__ == "__main__": # 载入预处理后的数据集 dataset = np.load('datasets.npz') X_train, y_train = dataset['X_train'], dataset['y_train'] X_val, y_val = dataset['X_val'], dataset['y_val'] input_shape = X_train.shape[1:] # 构建模型 model = build_model(input_shape) # 定义优化器和损失函数 optimizer = Adam(learning_rate=0.001) loss_weights = {'classification': 1., 'domain': 0.1} model.compile(optimizer=optimizer, loss={'classification': 'sparse_categorical_crossentropy', 'domain': DDCLoss()}, metrics=['accuracy'], loss_weights=loss_weights) # 模拟域标签（假设前半部分是源域，后半部分是目标域） batch_size = 32 steps_per_epoch = len(X_train) // batch_size def generate_batch(x, y): half_batch = batch_size // 2 while True: # 源域样本 indices_source = np.random.choice(len(y), size=half_batch) x_source = x[indices_source] y_class_source = y[indices_source] d_source = np.zeros((half_batch,)) # 目标域样本 indices_target = np.random.choice(len(y), size=batch_size - half_batch) x_target = x[indices_target] y_class_target = y[indices_target] d_target = np.ones((batch_size - half_batch,)) # 合并批次 x_batch = np.vstack([x_source, x_target]) y_class_batch = np.hstack([y_class_source, y_class_target]) d_batch = np.hstack([d_source, d_target]) yield ({'input_1': x_batch}, {'classification': y_class_batch, 'domain': d_batch}) generator = generate_batch(X_train, y_train) # 训练模型 model.fit(generator, epochs=10, steps_per_epoch=steps_per_epoch, validation_data=(X_val, {"classification": y_val}), verbose=1) # 输出评估结果 scores = model.evaluate(X_val, {"classification": y_val}) print(f"\nValidation Loss: {scores[0]}, Classification Accuracy: {scores[1]}") ``` --- ### 结果说明 - 上述两份代码分别完成了数据准备 (`data_preparation.py`) 和深度域混淆算法的应用 (`deep_domain_confusion.py`)。 - 这种设计保证了模块化开发的理念，便于维护和复用。 ---