layer弹窗一闪而过原因

import pickle import numpy as np import matplotlib.pyplot as plt import seaborn as sns import pandas as pd import os from sklearn.metrics import confusion_matrix from sklearn.preprocessing import LabelEncoder from sklearn.model_selection import train_test_split import torch import torch.nn as nn from torch.utils.data import TensorDataset, DataLoader from tqdm import tqdm # 设置随机种子 torch.manual_seed(42) np.random.seed(42) def load_data(file_path): """安全加载并预处理数据""" with open(file_path, 'rb') as f: data = pickle.load(f, encoding='latin1') # 调试信息 print(f"成功加载数据，总样本数: {len(data)}") sample_shape = next(iter(data.values())).shape print(f"单个样本形状: {sample_shape} (格式: [时间步长, 通道数, 特征数])") # 统一处理所有样本为相同长度（取最小长度） min_length = min(arr.shape[0] for arr in data.values()) print(f"统一长度: {min_length} (取所有样本的最小长度)") # 提取I路信号并统一长度 X = np.array([arr[:min_length, 0, 0] for arr in data.values()], dtype=np.float32) X = X.reshape(-1, min_length, 1) # 形状: (样本数, 时间步长, 特征数) # 提取调制类型标签 modulations = [key[0] for key in data.keys()] label_encoder = LabelEncoder() y = label_encoder.fit_transform(modulations) print(f"调制类型及对应标签:\n{pd.Series(label_encoder.classes_).to_frame('标签')}") # 分割数据集 X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, random_state=42) # 归一化 mean, std = X_train.mean(), X_train.std() X_train = (X_train - mean) / (std + 1e-8) X_test = (X_test - mean) / (std + 1e-8) return X_train, y_train, X_test, y_test, label_encoder class SignalTransformer(nn.Module): """稳健的Transformer模型""" def __init__(self, input_dim=1, num_classes=11, d_model=128, nhead=8, num_layers=3, dropout=0.1): super().__init__() self.input_proj = nn.Linear(input_dim, d_model) self.pos_encoder = nn.Parameter(torch.zeros(1, 1000, d_model)) encoder_layer = nn.TransformerEncoderLayer( d_model=d_model, nhead=nhead, dim_feedforward=512, dropout=dropout, batch_first=True) self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=num_layers) self.classifier = nn.Sequential( nn.LayerNorm(d_model), nn.Dropout(dropout), nn.Linear(d_model, num_classes) ) nn.init.xavier_uniform_(self.pos_encoder) def forward(self, x): x = self.input_proj(x) seq_len = x.size(1) x = x + self.pos_encoder[:, :seq_len, :] x = self.transformer(x).mean(dim=1) return self.classifier(x) def train_model(): # 设备配置 device = torch.device("cuda" if torch.cuda.is_available() else "cpu") print(f"\n训练设备: {device}") # 加载数据 data_path = r"D:\Anaconda\envs\DL\Lib\site-packages\caffe2\python\data\train\2016.04C.multisnr.pkl" try: X_train, y_train, X_test, y_test, label_encoder = load_data(data_path) except Exception as e: print(f"数据加载失败: {str(e)}") return # 转换为PyTorch张量 train_data = TensorDataset( torch.from_numpy(X_train).float(), torch.from_numpy(y_train).long() ) test_data = TensorDataset( torch.from_numpy(X_test).float(), torch.from_numpy(y_test).long() ) # 数据加载器 (batch size改回32) train_loader = DataLoader(train_data, batch_size=32, shuffle=True) test_loader = DataLoader(test_data, batch_size=32) # 模型初始化 model = SignalTransformer(num_classes=len(label_encoder.classes_)).to(device) criterion = nn.CrossEntropyLoss() optimizer = torch.optim.AdamW(model.parameters(), lr=3e-4, weight_decay=1e-4) # 训练记录 history = { 'epoch': [], 'train_loss': [], 'val_loss': [], 'val_acc': [], 'best_epoch': 0, 'best_acc': 0 } # 训练循环（保持5轮不变） for epoch in range(5): model.train() train_loss = 0 # 使用with语句管理进度条 with tqdm(train_loader, desc=f'Epoch {epoch+1}/5') as progress_bar: for inputs, labels in progress_bar: inputs, labels = inputs.to(device), labels.to(device) optimizer.zero_grad() outputs = model(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() train_loss += loss.item() progress_bar.set_postfix({'loss': loss.item()}) # 验证 model.eval() val_loss, correct = 0, 0 all_preds, all_labels = [], [] # 验证阶段也添加进度条 with tqdm(test_loader, desc='验证中') as val_bar: with torch.no_grad(): for inputs, labels in val_bar: inputs, labels = inputs.to(device), labels.to(device) outputs = model(inputs) val_loss += criterion(outputs, labels).item() preds = outputs.argmax(dim=1) correct += (preds == labels).sum().item() all_preds.extend(preds.cpu().numpy()) all_labels.extend(labels.cpu().numpy()) val_bar.set_postfix({'val_loss': val_loss/len(val_bar)}) # 记录结果 val_acc = correct / len(test_data) history['epoch'].append(epoch+1) history['train_loss'].append(train_loss/len(train_loader)) history['val_loss'].append(val_loss/len(test_loader)) history['val_acc'].append(val_acc) if val_acc > history['best_acc']: history['best_acc'] = val_acc history['best_epoch'] = epoch+1 torch.save(model.state_dict(), 'best_model.pth') # 使用tqdm.write确保输出不干扰进度条 tqdm.write(f"Epoch {epoch+1}: 训练损失={history['train_loss'][-1]:.4f}, " f"验证损失={history['val_loss'][-1]:.4f}, " f"准确率={val_acc:.4f}") # 保存结果 save_results(history, all_labels, all_preds, label_encoder.classes_) print(f"\n最佳模型在Epoch {history['best_epoch']}, 准确率={history['best_acc']:.4f}") def save_results(history, true_labels, pred_labels, classes): """保存所有结果并显示可视化""" os.makedirs('results', exist_ok=True) # 创建两个独立图形 fig1, ax1 = plt.subplots(figsize=(12, 10)) fig2, (ax2, ax3) = plt.subplots(1, 2, figsize=(12, 5)) # 1. 混淆矩阵 cm = confusion_matrix(true_labels, pred_labels) sns.heatmap(cm, annot=True, fmt='d', xticklabels=classes, yticklabels=classes, ax=ax1) ax1.set_title("Confusion Matrix") fig1.savefig("results/confusion_matrix.png", bbox_inches='tight') # 2. 训练曲线 ax2.plot(history['epoch'], history['train_loss'], label='Train Loss') ax2.plot(history['epoch'], history['val_loss'], label='Validation Loss') ax2.set_xlabel('Epoch') ax2.set_ylabel('Loss') ax2.legend() ax3.plot(history['epoch'], history['val_acc'], label='Validation Accuracy', color='green') ax3.set_xlabel('Epoch') ax3.set_ylabel('Accuracy') ax3.legend() fig2.savefig("results/training_metrics.png", bbox_inches='tight') # 同时显示两个窗口 plt.show(block=False) # 不阻塞代码执行 plt.pause(0.1) # 短暂暂停确保窗口创建 # 3. 训练日志表格 df = pd.DataFrame({ 'Epoch': history['epoch'], 'Training Loss': history['train_loss'], 'Validation Loss': history['val_loss'], 'Accuracy': history['val_acc'] }) df.to_csv("results/training_log.csv", index=False) print("\n训练日志摘要:") print(df.round(4)) if __name__ == "__main__": train_model() 以上代码输出的时候，混淆矩阵弹窗和损失精度曲线图弹窗一闪即逝，我需要这两个弹窗能正常显示，请你修改后发我完整代码