import pandas as pd import numpy as np import tensorflow as tf import matplotlib.pyplot as plt def data_read(data_address): df = pd.read_csv(data_address) label_mapping = {label: idx for idx, label in enumerate(df['Problem'].unique())} df['Problem'] = df['Problem'].map(label_mapping) df.replace([np.inf, -np.inf], np.nan, inplace=True) df.dropna(inplace=True) X = np.array(df['Voltage']) Y = np.array(df['Problem']) # 归一化处理 X = (X - np.min(X)) / (np.max(X) - np.min(X)) # 转换为时间序列数据格式 time_steps = 34 X_series, Y_series = [], [] for i in range(0, len(X) - time_steps): X_series.append(X[i:(i + time_steps)]) Y_series.append(Y[i + time_steps - 1]) return np.array(X_series).reshape(-1, time_steps, 1), np.array(Y_series) X_train, Y_train = data_read( 'Liu\data\VOLTAGE-QUALITY-CLASSIFICATION-MODEL--main\Voltage Quality.csv') X_test, Y_test = data_read( 'Liu/data/VOLTAGE-QUALITY-CLASSIFICATION-MODEL--main/Voltage Quality Test.csv') # 获取类别数量 n_classes = len(np.unique(Y_train)) # 构建模型 model = tf.keras.models.Sequential([ tf.keras.layers.LSTM(50, return_sequences=True, input_shape=(34, 1)), tf.keras.layers.LSTM(50), tf.keras.layers.Dense(n_classes, activation='softmax') # 修改为适应多分类 ]) # 编译模型 model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) # 训练模型 model.fit(X_train, Y_train, epochs=10, validation_split=0.2) # 评估模型 loss, accuracy = model.evaluate(X_test, Y_test) print(X_train.shape) # 制作扰动数据 # 损失函数 loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) # 定义损失函数 loss_fn = tf.keras.losses.SparseCategoricalCrossentropy() # 转换X_train和Y_train为TensorFlow张量 X_train_tensor = tf.convert_to_tensor(X_train, dtype=tf.float32) Y_train_tensor = tf.convert_to_tensor(Y_train, dtype=tf.int32) # 使用tf.GradientTape来计算梯度 with tf.GradientTape() as tape: # 确保tape监视输入张量 tape.watch(X_train_tensor) # 前向传播，计算预测值 predictions = model(X_train_tensor) # 计算损失 loss = loss_fn(Y_train_tensor, predictions) # 计算关于输入X的梯度 gradients = tape.gradient(loss, X_train_tensor) # 计算每个输入的梯度的L2范数 gradient_magnitudes = tf.norm(gradients, axis=1) # 创建每个gamma对应的准确率的字典 accuracy_per_gamma = {} # 平坦化梯度 flattened_gradients = tf.reshape(gradients, [-1]) # 选择最大的γ * |X|个梯度 for gamma in [0.05, 0.1, 0.2, 0.4]: num_gradients_to_select = int(gamma * tf.size(flattened_gradients, out_type=tf.dtypes.float32)) top_gradients_indices = tf.argsort(flattened_gradients, direction='DESCENDING')[:num_gradients_to_select] # 创建一个新的梯度张量，初始化为原始梯度的副本 updated_gradients = tf.identity(flattened_gradients) # 创建一个布尔掩码，其中选定的最大梯度为False，其他为True mask = tf.ones_like(updated_gradients, dtype=bool) mask = tf.tensor_scatter_nd_update(mask, tf.expand_dims(top_gradients_indices, 1), tf.zeros_like(top_gradients_indices, dtype=bool)) # 使用这个掩码更新梯度 updated_gradients = tf.where(mask, tf.zeros_like(updated_gradients), updated_gradients) # 将梯度重构为原始形状 updated_gradients = tf.reshape(updated_gradients, tf.shape(gradients)) # 创建准确率列表 accuracy_list = [] for learning_rate in [0.1, 0.2, 0.3, 0.4, 0.5]: # 应用学习率到梯度 scaled_gradients = learning_rate * updated_gradients # 使用缩放后的梯度更新X_train_tensor X_train_updated = X_train_tensor - scaled_gradients tf.reshape(X_train_updated, (3332,34,1)) X_train_updated = X_train_updated.numpy() # 编译模型 model.compile( optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) # 训练模型 model.fit(X_train_updated, Y_train, epochs=1500, validation_split=0.2) # 评估模型 loss, accuracy = model.evaluate(X_test, Y_test) # 记录准确率 accuracy_list.append(accuracy) # 记录该gamma下的准确率 accuracy_per_gamma[gamma] = accuracy_list # 学习率样本 learning_rates = [0.1, 0.2, 0.3, 0.4, 0.5] # 不同的gamma值 gammas = [0.05, 0.1, 0.2, 0.4] # 创建图像 plt.figure(figsize=(10, 6)) # 为每个gamma值绘制曲线 for gamma in gammas: plt.plot(learning_rates, accuracy_per_gamma[gamma], marker='o', label=f'Gamma={gamma}') # 添加标题和标签 plt.title('Accuracy vs Learning Rate for Different Gammas') plt.xlabel('Learning Rate') plt.ylabel('Accuracy') plt.legend() # 显示图像 plt.show()