``` if __name__ == '__main__': # 载入训练集和验证集的数据 x_train, y_train = load_dataset('./划分后的数据/训练集') x_val, _ = load_dataset('./划分后的数据/验证集') input_dim = x_train.shape[1] # 创建共享特征提取器 feature_extractor = Sequential([ Dense(128, activation='relu', input_shape=(input_dim,)), Dropout(0.5), Dense(64, activation='relu') ]) # 创建分类器 classifier_inputs = Input(shape=(64,)) classification_output = Dense(1, activation='sigmoid')(classifier_inputs) classifier_model = Model(inputs=classifier_inputs, outputs=classification_output) classifier_model.compile(optimizer=Adam(), loss='binary_crossentropy', metrics=['accuracy']) # 创建域判别器 discriminator_inputs = Input(shape=(64,)) domain_output = Dense(1, activation='sigmoid')(discriminator_inputs) discriminator_model = Model(inputs=discriminator_inputs, outputs=domain_output) discriminator_model.compile(optimizer=Adam(), loss='binary_crossentropy', metrics=['accuracy']) # 封装整体网络结构用于反向传播 combined_input = feature_extractor.output classification_output_combined = classifier_model(combined_input) domain_output_combined = discriminator_model(combined_input) combined_model = Model(feature_extractor.input, [classification_output_combined, domain_output_combined]) # 冻结判别器权重以仅更新分类器权重 discriminator_model.trainable = False combined_model.compile( optimizer=Adam(), loss=['binary_crossentropy', 'binary_crossentropy'], metrics=['accuracy'] ) # 准备域标签 (源域 vs 目标域) batch_size = 32 num_batches = len(x_train) // batch_size for epoch in range(10): # 设置迭代次数 indices = np.arange(len(x_train)) np.random.shuffle(indices) for i in range(num_batches): idx = indices[i * batch_size:(i + 1) * batch_size] source_batch = x_train[idx] # 源域样本 target_batch = x_val[np.random.choice(len(x_val), size=batch_size)] # 随机采样目标域样本 X_combined = np.vstack([source_batch, target_batch]) Y_domain_labels = np.array([1] * batch_size + [0] * batch_size) # 源域标记为1，目标域标记为0 Y_class_labels = y_train[idx].astype(float) # 只有源域才有类别标签 # 更新特征提取器+分类器 combined_loss = combined_model.train_on_batch(X_combined, [Y_class_labels, Y_domain_labels]) # 解冻判别器并单独训练它 discriminator_model.trainable = True disc_loss = discriminator_model.train_on_batch(feature_extractor.predict(source_batch), np.ones(batch_size)) # 判别器区分真实分布 disc_loss += discriminator_model.train_on_batch(feature_extractor.predict(target_batch), np.zeros(batch_size)) discriminator_model.trainable = False print(f"Epoch {epoch + 1}: Combined Loss={combined_loss}, Discriminator Loss={disc_loss}") print("Domain adaptation training completed.") # 测试阶段 - 使用验证集评估分类精度 extracted_features = feature_extractor.predict(x_val) predictions = classifier_model.predict(extracted_features) accuracy = np.mean((predictions > 0.5).flatten() == _) # 替换_为目标验证集的真实标签 print(f"Validation Accuracy: {accuracy:.4f}") validation_accuracy = combined_model.evaluate(x_val, [y_val, ...], verbose=0)[1] # 获取第二个输出的acc指标 print(f"Final Validation Classification Accuracy: {validation_accuracy:.4f}")```运行结果：2025-03-29 15:32:12.506410: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX AVX2 To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags. Traceback (most recent call last): File "C:/Users/Lenovo/AppData/Roaming/JetBrains/PyCharmCE2020.2/scratches/scratch_19.py", line 104, in <module> Y_class_labels = y_train[idx].astype(float) # 只有源域才有类别标签 IndexError: index 121655 is out of bounds for axis 0 with size 288