初步完成任务二Task2

.devcontainer/cpu/Dockerfile

@@ -1,23 +1,6 @@
-#FROM elainasuki/ros:ros2-humble-full-v3
-FROM elainasuki/ros:ros2-humble-full-0614
-
-ARG USERNAME=Elaina
-
-ARG USER_GID=$USER_UID
-
-ARG GROUP_NAME=wheel
-RUN apt-get update \
-    && apt-get install -y  sudo vim nautilus 
-USER ${USERNAME}
-#安装前置依赖
+FROM ubuntu:22.04
+RUN apt-get update\ 
+    && apt-get install -y git  python3 python3-pip 
 RUN pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
 
-# 安装yolo
-RUN pip install imutils ultralytics openvino
-#安装其他依赖
-
-RUN pip install jupyter d2l==0.17.6   
-
-USER root
-RUN  echo "source /opt/ros/humble/setup.bash" >> /home/${USERNAME}/.bashrc
-ENV PYTHONPATH="/home/Elaina/yolo:${PYTHONPATH}"
+RUN pip install pandas openpyxl matplotlib

.devcontainer/cpu/build.bash

@@ -9,7 +9,7 @@ echo "脚本目录: $SCRIPT_DIR"
 echo "父目录: $PARENT_DIR"
 
 # 设置默认 tag
-TAG="pytorch"
+TAG="pytorch_cpu"
 
 # 从外部传入的 IMAGE_REPO（格式：ghcr.io/user/repo 或 docker.io/user/repo）
 IMAGE_REPO="elainasuki/other"

.devcontainer/devcontainer.json

@@ -2,7 +2,7 @@
     "name": "information-task-container",
     "dockerComposeFile": "docker-compose.yml",
     "service": "gpu-service",
-    "workspaceFolder": "/home/Elaina/pytorch",
+    "workspaceFolder": "/pytorch",
     "shutdownAction": "stopCompose",
     "customizations": {
         "vscode": {

.devcontainer/docker-compose.yml

@@ -1,10 +1,7 @@
 version: '3'
 services:
   pytorch-service:
-    # build:
-    #   context: ..
-    #   dockerfile: .devcontainer/Dockerfile
-    image: elainasuki/other:pytorch
+    image: elainasuki/other:pytorch_cpu
     container_name: information-cpu-container
     environment:
       - DISPLAY=${DISPLAY}
@@ -14,21 +11,13 @@ services:
       - TERM=xterm-256color
     volumes:
       - /tmp/.X11-unix:/tmp/.X11-unix
-      - ./..:/home/Elaina/pytorch
-      - /dev:/dev
-    network_mode: host
-    pid: "host" # 添加 pid 命名空间共享
-    ipc: "host" # 添加 ipc 命名空间共享    
-    privileged: true
+      - ./..:/pytorch/
+    entrypoint: [ '/bin/bash' ]
     stdin_open: true
     tty: true
-    user: "Elaina"
     # runtime: "nvidia"
-    working_dir: "/home/Elaina/pytorch" # 指定默认工作目录
+    working_dir: "/pytorch" # 指定默认工作目录
   gpu-service:
-    # build:
-    #   context: ..
-    #   dockerfile: .devcontainer/Dockerfile
     image: elainasuki/other:pytorch_gpu
     container_name: information-task-container
     environment:
@@ -39,14 +28,9 @@ services:
       - TERM=xterm-256color
     volumes:
       - /tmp/.X11-unix:/tmp/.X11-unix
-      - ./..:/home/Elaina/pytorch
-      - /dev:/dev
-    network_mode: host
-    pid: "host" # 添加 pid 命名空间共享
-    ipc: "host" # 添加 ipc 命名空间共享    
-    privileged: true
-    stdin_open: true
-    tty: true
-    user: "Elaina"
+      - ./..:/pytorch
     runtime: "nvidia"
-    working_dir: "/home/Elaina/pytorch" # 指定默认工作目录
+    tty: true
+    stdin_open: true
+    entrypoint: [ '/bin/bash' ]
+    working_dir: "/pytorch" # 指定默认工作目录

.devcontainer/gpu/Dockerfile

@@ -1,44 +1,6 @@
-# 基础镜像：ROS 2 Humble（官方桌面版，基于Ubuntu 22.04）
-FROM elainasuki/ros:ros2-humble-full-0614
+FROM pytorch/pytorch:2.9.0-cuda13.0-cudnn9-runtime
 
-# 修复未定义变量
-ARG USERNAME=Elaina
-ARG USER_UID=1000
-ARG USER_GID=$USER_UID
+RUN pip3 install pandas openpyxl matplotlib
 
-# 设置非交互模式+pip缓存目录（避免占用根目录空间）
-ENV DEBIAN_FRONTEND=noninteractive
-ENV PIP_CACHE_DIR=/tmp/pip-cache
-
-# 1. 精简基础依赖（仅保留必需项）+ 清理缓存
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    wget \
-    ca-certificates \
-    python3-pip \
-    && apt-get clean \
-    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
-
-# 2. 仅安装CUDA运行时（极致精简）
-RUN wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb \
-    && dpkg -i cuda-keyring_1.1-1_all.deb \
-    && apt-get update \
-    && apt-get install -y --no-install-recommends \
-    cuda-runtime-12-1 \
-    libcudnn8=8.9.2.26-1+cuda12.1 \
-    && apt-get clean \
-    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
-    && rm -f cuda-keyring_1.1-1_all.deb
-
-# 3. 配置环境变量（修复未定义问题）
-ENV PATH=/usr/local/cuda-12.1/bin:${PATH}
-ENV LD_LIBRARY_PATH=/usr/local/cuda-12.1/lib64:${LD_LIBRARY_PATH:-/usr/local/lib}
-
-# 4. 优化PyTorch安装（节省空间）
-RUN pip3 install --no-cache-dir --upgrade pip \
-    && pip3 install --no-cache-dir --ignore-installed sympy \
-    && pip3 install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 \
-    && rm -rf $PIP_CACHE_DIR /tmp/* /var/tmp/*
-#安装pandas 和openpyxl
-RUN pip3 install pandas openpyxl
-# 5. 配置ROS 2环境
-RUN echo "source /opt/ros/humble/setup.bash" >> ~/.bashrc
+RUN apt-get update &&\
+    apt-get install -y git 

.gitignore

@@ -0,0 +1,4 @@
+.~*
+model/
+*.png
+*.pyc

.vscode/settings.json

@@ -0,0 +1,3 @@
+{
+    "python.analysis.typeCheckingMode": "standard"
+}

src/infer.py

@@ -0,0 +1,25 @@
+from visual import *
+from model import *
+from load_data import *
+def infer():
+    #初始化参数
+    model_path='/pytorch/model/best_transformer.pth'
+    device=torch.device('cuda' if torch.cuda.is_available() else'cpu')
+    #加载模型
+    model=MLPModel(input_size=8,hidden_size=24,output_size=2,device=device)
+    model.load_state_dict(torch.load(model_path, map_location=device, weights_only=True)['model_state_dict'])
+    #加载数据
+    data_path='/pytorch/data/task2.xlsx'
+    batcher = DataBatcher(file_path=data_path, val_ratio=0.2, batch_size=16,device=device)
+    val_inputs, val_outputs = batcher.getValBatches()
+    #进行推理
+    output=model(val_inputs)
+    #取第1batch的结果比较
+    print("真实值:", val_outputs[0])
+    print("预测值:", output[0])
+    # 计算均方误差
+    loss_fn=torch.nn.MSELoss()
+    loss=loss_fn(output,val_outputs)
+    print(f"验证集均方误差: {loss.item():.4f}")
+if __name__ == "__main__":
+    infer()

src/load_data.py

@@ -17,13 +17,13 @@ def LoadTask2Data(file_path)->tuple[torch.Tensor, torch.Tensor]:
     outputs_tensor = torch.tensor(outputs, dtype=torch.float32)
     return inputs_tensor, outputs_tensor
 class DataBatcher:
-    def __init__(self, file_path: str, val_ratio: float = 0.2, batch_size: int = 32, shuffle: bool = True):
+    def __init__(self, file_path: str, val_ratio: float = 0.2, batch_size: int = 32, shuffle: bool = True,device: torch.device = torch.device('cpu')):
         """
         数据批处理工具类，支持划分验证集和生成批次数据
         """
         self.batch_size = batch_size
         self.shuffle = shuffle
-
+        self.device = device
         # 加载原始数据
         self.inputs, self.outputs = LoadTask2Data(file_path)
 
@@ -81,16 +81,16 @@ def _splitAndCreateBatches(self, val_ratio: float):
 
     def getTrainBatches(self):
         """获取训练集批次张量 (输入, 输出)"""
-        return self.train_inputs, self.train_outputs
+        return self.train_inputs.to(self.device), self.train_outputs.to(self.device)
 
     def getValBatches(self):
         """获取验证集批次张量 (输入, 输出)"""
-        return self.val_inputs, self.val_outputs
+        return self.val_inputs.to(self.device), self.val_outputs.to(self.device)
 
 
 if __name__ == "__main__":
     # 测试代码
-    file_path = os.path.join("/home/Elaina/pytorch/data", "task2.xlsx")
+    file_path = os.path.join("/pytorch/data", "task2.xlsx")
 
     # 初始化批处理工具
     batcher = DataBatcher(file_path, val_ratio=0.2, batch_size=16)    

src/model.py

@@ -0,0 +1,19 @@
+import torch.nn as nn
+import os,torch
+class MLPModel(nn.Module):
+    """简单的多层感知机模型"""
+
+    def __init__(self, input_size, hidden_size, output_size, device=torch.device('cpu')):
+        super().__init__()
+        self.linear1 = nn.Linear(input_size, hidden_size)
+        self.relu = nn.ReLU()
+        self.linear2 = nn.Linear(hidden_size, output_size)
+        self.device=device
+        self.to(device)  # 关键：初始化时就把模型移到目标设备
+    def forward(self, x:torch.Tensor)->torch.Tensor:
+        assert isinstance(x, torch.Tensor),  "输入必须是torch.Tensor类型"
+        x = x.to(self.device)  # 确保输入数据在正确的设备上
+        x = self.linear1(x)
+        x = self.relu(x)
+        x = self.linear2(x)
+        return x

src/train.py

@@ -0,0 +1,32 @@
+from visual import *
+from model import *
+from load_data import *
+def train():
+    device=torch.device('cuda' if torch.cuda.is_available() else'cpu')
+    model=MLPModel(input_size=8,hidden_size=24,output_size=2,device=device)
+    data_path='/pytorch/data/task2.xlsx'
+    model_path='/pytorch/model/'
+    batcher = DataBatcher(file_path=data_path, val_ratio=0.2, batch_size=16,device=device)
+    train_inputs, train_outputs = batcher.getTrainBatches()
+    val_inputs, val_outputs = batcher.getValBatches()
+    #平方损失
+    loss_fn=torch.nn.MSELoss()
+    optimizer=torch.optim.Adam(model.parameters(),lr=0.001)
+
+    visual=SaveAndVisual(model_dir=model_path, loss_img_path=model_path+'loss_curve.png')
+    num_epoch=600
+    visual.loadModel(model,optimizer,device)
+    for epoch in range(num_epoch):
+        model.train()
+        inputs=train_inputs
+        targets=train_outputs
+        optimizer.zero_grad()
+        outputs=model(inputs)
+        loss=loss_fn(outputs,targets)
+        loss.backward()
+        optimizer.step()
+        visual.updateVisualization(epoch,loss.item())
+    visual.finalizeVisualization()
+
+if __name__ == "__main__":
+    train()

src/visual.py

@@ -0,0 +1,86 @@
+# visual.py
+import matplotlib.pyplot as plt
+import os,torch
+class SaveAndVisual:
+    """模型管理类，负责模型保存和训练可视化"""
+
+    def __init__(self, model_dir='models', loss_img_path='loss_curve.png'):
+        self.model_dir = model_dir
+        self.loss_img_path = loss_img_path
+        self.epoch_losses = []  # 存储每个epoch的损失
+        self.epoch_indices = []  # 存储epoch索引
+        self._init_visualization()
+        self._init_model_dir()
+        self.loop_count=0
+    def _init_model_dir(self):
+        """初始化模型保存目录"""
+        os.makedirs(self.model_dir, exist_ok=True)
+
+    def _init_visualization(self):
+        """初始化可视化环境"""
+        plt.ion()  # 开启交互模式
+        self.fig, self.ax = plt.subplots(figsize=(10, 6))
+        self.ax.set_xlabel("Epoch")
+        self.ax.set_ylabel("Loss")
+        self.ax.set_title("Training Loss (per Epoch)")
+        self.line, = self.ax.plot([], [], label="Epoch Loss")
+        self.ax.legend()
+        self.min_loss=float('inf')
+    def loadModel(self, model:torch.nn.Module, optimizer, device):
+        """加载已保存的模型"""
+        model_path = os.path.join(self.model_dir, 'best_transformer.pth')
+        self.model=model
+        self.optimizer=optimizer
+        if os.path.exists(model_path):
+            checkpoint = torch.load(model_path, map_location=device, weights_only=True)
+            model.load_state_dict(checkpoint['model_state_dict'])
+            optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
+            print(f"发现现有模型，其损失为: {checkpoint['loss']:.4f}")
+            self.min_loss=checkpoint['loss']
+            return checkpoint['loss']
+        return float('inf')
+
+    def saveModel(self, model, optimizer, epoch, loss):
+        """保存模型检查点"""
+        model_path = os.path.join(self.model_dir, 'best_transformer.pth')
+        torch.save({
+            'epoch': epoch,
+            'model_state_dict': model.state_dict(),
+            'optimizer_state_dict': optimizer.state_dict(),
+            'loss': loss,
+        }, model_path)
+        # print(f"  保存最佳模型（损失：{loss:.4f})")
+
+    def updateVisualization(self, epoch, loss):
+        """更新训练损失可视化"""
+        self.epoch_losses.append(loss)
+        self.epoch_indices.append(epoch)
+        self.loop_count+=1
+        if(self.loop_count%10==0):
+            self.loop_count=0
+            print(f"  训练损失（第{epoch+1}轮）：{loss:.4f}")
+        # 更新图像数据
+        self.line.set_data(self.epoch_indices, self.epoch_losses)
+        self.ax.relim()  # 重新计算坐标轴范围
+        self.ax.autoscale_view()  # 自动调整视图
+        if(loss<self.min_loss ):
+            self.min_loss=loss
+            self.saveModel(self.model,self.optimizer,epoch,loss)
+        plt.draw()
+        plt.pause(0.01)
+
+    def finalizeVisualization(self):
+        """训练结束后保存并显示最终图像"""
+        plt.ioff()  # 关闭交互模式
+        self.ax.set_title("Training Loss (Final)")
+        plt.savefig(self.loss_img_path)
+        plt.show()
+if __name__ == "__main__":
+    plt.figure(figsize=(6, 4))
+    plt.plot([1, 2, 3, 4], [1, 4, 9, 16], 'r-', label='test')
+    plt.xlabel('X')
+    plt.ylabel('Y')
+    plt.title('GUI')
+    plt.legend()
+    plt.grid(True)
+    plt.show()  # 弹出窗口显示测试图