plot_loss.py

import matplotlib.pyplot as plt
import csv
import sys
import os

def plot_loss(filename="loss_log.txt"):
    if not os.path.exists(filename):
        print(f"错误: 找不到文件 {filename}。请先运行 C++ 程序进行训练。")
        return

    epochs = []
    losses = []

    print(f"正在读取 {filename} ...")

    try:
        with open(filename, 'r') as f:
            reader = csv.reader(f)
            next(reader)  # 跳过第一行表头 (epoch,loss)
            
            for row in reader:
                if len(row) >= 2:
                    epochs.append(int(row[0]))
                    losses.append(float(row[1]))
    except Exception as e:
        print(f"读取文件出错: {e}")
        return

    # 开始画图
    plt.figure(figsize=(10, 6))
    plt.plot(epochs, losses, label='Training Loss', color='blue', linewidth=2)
    
    # 添加标题和标签
    plt.title('Neural Network Training Loss Curve', fontsize=16)
    plt.xlabel('Epoch', fontsize=12)
    plt.ylabel('Average MSE Loss', fontsize=12)
    plt.grid(True, which='both', linestyle='--', alpha=0.7)
    plt.legend()

    # 标记最低点
    if losses:
        min_loss = min(losses)
        min_epoch = epochs[losses.index(min_loss)]
        plt.annotate(f'Min Loss: {min_loss:.6f}', 
                     xy=(min_epoch, min_loss), 
                     xytext=(min_epoch, min_loss + 0.05),
                     arrowprops=dict(facecolor='red', shrink=0.05))

    print("正在显示图表...")
    plt.show()

if __name__ == "__main__":
    # 如果命令行传了参数，就用参数指定的文件名，否则默认 loss_log.txt
    log_file = sys.argv[1] if len(sys.argv) > 1 else "loss_log.txt"
    plot_loss(log_file)