大模型算法学习2026.6.13

「維他檸檬茶」2026-06-14 12:55

这两天看完了RNN、GRU和LSTM，下面给出一个使用PyTorch实现的、不同神经网络实现简单直观的项目，用于MNIST手写数字识别，同样作为一个深度学习模板进行学习记忆。

RNN:

复制代码 
import os
# 设置 PyTorch 数据集国内镜像
os.environ["TORCHVISION_DATASET_MIRROR"] = "https://mirrors.tuna.tsinghua.edu.cn/pytorch-datasets"

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
import matplotlib.pyplot as plt

# 1. 设置设备
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"使用设备: {device}")

# 2. 超参数
input_size = 28
sequence_length = 28
hidden_size = 128
num_classes = 10
num_epochs = 5
batch_size = 64
learning_rate = 0.001

# 3. 数据预处理
transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Lambda(lambda x: x.squeeze(0))
])

# 已有数据集后把 download=True 改为 download=False
train_dataset = datasets.MNIST(root='./data', train=True, transform=transform, download=True)
test_dataset  = datasets.MNIST(root='./data', train=False, transform=transform, download=True)

train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
test_loader  = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)

class SimpleLSTM(nn.Module):
    def __init__(self, input_size, hidden_size, num_classes):
        super().__init__()
        self.hidden_size = hidden_size
        self.lstm = nn.RNN(input_size, hidden_size, batch_first=True)
        self.fc = nn.Linear(hidden_size, num_classes)

    def forward(self, x):
        batch = x.size(0)
        # LSTM 需要 h0、c0 两个初始状态
        h0 = torch.zeros(1, batch, self.hidden_size).to(x.device)
        # c0 = torch.zeros(1, batch, self.hidden_size).to(x.device)
        out, _ = self.lstm(x, h0)
        # 取最后一个时间步输出
        out = out[:, -1, :]
        out = self.fc(out)
        return out

model = SimpleLSTM(input_size, hidden_size, num_classes).to(device)

# 5. 损失函数、优化器
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

# 6. 训练循环
print("开始训练...")
train_losses = []
for epoch in range(num_epochs):
    model.train()
    total_loss = 0.0
    for images, labels in train_loader:
        images = images.to(device)
        labels = labels.to(device)

        outputs = model(images)
        loss = criterion(outputs, labels)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_loss += loss.item()

    avg_loss = total_loss / len(train_loader)
    train_losses.append(avg_loss)
    print(f"Epoch [{epoch+1}/{num_epochs}], Loss: {avg_loss:.4f}")

# 7. 测试评估
model.eval()
correct = 0
total = 0
with torch.no_grad():
    for images, labels in test_loader:
        images = images.to(device)
        labels = labels.to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

print(f"\n测试集准确率: {100 * correct / total:.2f}%")

# 8. 可视化单张图片
first_image = test_dataset[0][0].cpu().numpy()
first_label = test_dataset[0][1]
model.eval()
with torch.no_grad():
    pred = model(test_dataset[0][0].unsqueeze(0).to(device))
    pred_class = torch.argmax(pred, dim=1).item()

plt.imshow(first_image, cmap='gray')
plt.title(f"真实标签: {first_label}, 预测结果: {pred_class}")
plt.show()

GRU:

复制代码 
import os
# 设置 PyTorch 数据集国内镜像
os.environ["TORCHVISION_DATASET_MIRROR"] = "https://mirrors.tuna.tsinghua.edu.cn/pytorch-datasets"

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
import matplotlib.pyplot as plt

# 1. 设置设备
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"使用设备: {device}")

# 2. 超参数
input_size = 28
sequence_length = 28
hidden_size = 128
num_classes = 10
num_epochs = 5
batch_size = 64
learning_rate = 0.001

# 3. 数据预处理
transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Lambda(lambda x: x.squeeze(0))
])

# 已有数据集后把 download=True 改为 download=False
train_dataset = datasets.MNIST(root='./data', train=True, transform=transform, download=True)
test_dataset  = datasets.MNIST(root='./data', train=False, transform=transform, download=True)

train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
test_loader  = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)

class SimpleLSTM(nn.Module):
    def __init__(self, input_size, hidden_size, num_classes):
        super().__init__()
        self.hidden_size = hidden_size
        self.lstm = nn.GRU(input_size, hidden_size, batch_first=True)
        self.fc = nn.Linear(hidden_size, num_classes)

    def forward(self, x):
        batch = x.size(0)
        # LSTM 需要 h0、c0 两个初始状态
        h0 = torch.zeros(1, batch, self.hidden_size).to(x.device)
        # c0 = torch.zeros(1, batch, self.hidden_size).to(x.device)
        out, _ = self.lstm(x, h0)
        # 取最后一个时间步输出
        out = out[:, -1, :]
        out = self.fc(out)
        return out

model = SimpleLSTM(input_size, hidden_size, num_classes).to(device)

# 5. 损失函数、优化器
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

# 6. 训练循环
print("开始训练...")
train_losses = []
for epoch in range(num_epochs):
    model.train()
    total_loss = 0.0
    for images, labels in train_loader:
        images = images.to(device)
        labels = labels.to(device)

        outputs = model(images)
        loss = criterion(outputs, labels)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_loss += loss.item()

    avg_loss = total_loss / len(train_loader)
    train_losses.append(avg_loss)
    print(f"Epoch [{epoch+1}/{num_epochs}], Loss: {avg_loss:.4f}")

# 7. 测试评估
model.eval()
correct = 0
total = 0
with torch.no_grad():
    for images, labels in test_loader:
        images = images.to(device)
        labels = labels.to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

print(f"\n测试集准确率: {100 * correct / total:.2f}%")

# 8. 可视化单张图片
first_image = test_dataset[0][0].cpu().numpy()
first_label = test_dataset[0][1]
model.eval()
with torch.no_grad():
    pred = model(test_dataset[0][0].unsqueeze(0).to(device))
    pred_class = torch.argmax(pred, dim=1).item()

plt.imshow(first_image, cmap='gray')
plt.title(f"真实标签: {first_label}, 预测结果: {pred_class}")
plt.show()

LSTM:

复制代码 
import os
# 设置 PyTorch 数据集国内镜像
os.environ["TORCHVISION_DATASET_MIRROR"] = "https://mirrors.tuna.tsinghua.edu.cn/pytorch-datasets"

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
import matplotlib.pyplot as plt

# 1. 设置设备
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"使用设备: {device}")

# 2. 超参数
input_size = 28
sequence_length = 28
hidden_size = 128
num_classes = 10
num_epochs = 5
batch_size = 64
learning_rate = 0.001

# 3. 数据预处理
transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Lambda(lambda x: x.squeeze(0))
])

# 已有数据集后把 download=True 改为 download=False
train_dataset = datasets.MNIST(root='./data', train=True, transform=transform, download=True)
test_dataset  = datasets.MNIST(root='./data', train=False, transform=transform, download=True)

train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
test_loader  = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)

# 4. LSTM 模型（修复双状态问题）
class SimpleLSTM(nn.Module):
    def __init__(self, input_size, hidden_size, num_classes):
        super().__init__()
        self.hidden_size = hidden_size
        # LSTM 不要加 nonlinearity 参数（LSTM 内部固定激活）
        self.lstm = nn.LSTM(input_size, hidden_size, batch_first=True)
        self.fc = nn.Linear(hidden_size, num_classes)

    def forward(self, x):
        batch = x.size(0)
        # LSTM 需要 h0、c0 两个初始状态
        h0 = torch.zeros(1, batch, self.hidden_size).to(x.device)
        c0 = torch.zeros(1, batch, self.hidden_size).to(x.device)
        out, _ = self.lstm(x, (h0,c0))
        # 取最后一个时间步输出
        out = out[:, -1, :]
        out = self.fc(out)
        return out

model = SimpleLSTM(input_size, hidden_size, num_classes).to(device)

# 5. 损失函数、优化器
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

# 6. 训练循环
print("开始训练...")
train_losses = []
for epoch in range(num_epochs):
    model.train()
    total_loss = 0.0
    for images, labels in train_loader:
        images = images.to(device)
        labels = labels.to(device)

        outputs = model(images)
        loss = criterion(outputs, labels)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_loss += loss.item()

    avg_loss = total_loss / len(train_loader)
    train_losses.append(avg_loss)
    print(f"Epoch [{epoch+1}/{num_epochs}], Loss: {avg_loss:.4f}")

# 7. 测试评估
model.eval()
correct = 0
total = 0
with torch.no_grad():
    for images, labels in test_loader:
        images = images.to(device)
        labels = labels.to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

print(f"\n测试集准确率: {100 * correct / total:.2f}%")

# 8. 可视化单张图片
first_image = test_dataset[0][0].cpu().numpy()
first_label = test_dataset[0][1]
model.eval()
with torch.no_grad():
    pred = model(test_dataset[0][0].unsqueeze(0).to(device))
    pred_class = torch.argmax(pred, dim=1).item()

plt.imshow(first_image, cmap='gray')
plt.title(f"真实标签: {first_label}, 预测结果: {pred_class}")
plt.show()
相关推荐
机器人零零壹1 小时前
技术白皮书：iRobotCAM如何解决机器人激光切割应用中的核心难点，从CAD离线编程到双目视觉定位与轨迹自适应提升复杂空间曲线切割的精度、效率与柔性化水平
人工智能·算法
ZhengEnCi4 小时前
大模型的分类到底怎么分：从基础模型到推理模型、从通用到专用的完整图谱
算法
前进的李工4 小时前
Agent 学习笔记 05：系统提示词应该怎么写
笔记·学习·prompt·agent·个人开发·few-shot·ai应用开发
取名真是5 小时前
Python 算法知识点从基础到进阶
开发语言·python·算法
罗西的思考5 小时前
【轻量化复刻大模型 / KIMI K3】KDA 模块交互式演示
人工智能·算法·机器学习
正在走向自律6 小时前
【3/5】人工智能训练师 三级/高级工 考证学习方向：从“独立执行”到“设计、归因与全链条负责”
人工智能·学习·人工智能训练师·注方案设计·数据质控负责人·模型评测与调优辅助·ai训练项目交付
nm_zfyjt7 小时前
信息流广告投放优化实践：线索成本评估模型与转化链路重构（家装建材案例）
算法·重构·内蒙古朋友圈广告服务商·内蒙古朋友圈获客
CCIE-Yasuo8 小时前
网络安全-学习数据库-Mysql-2
数据库·学习·mysql
La Pulga8 小时前
【FreeRTOS源码学习笔记】任务切换
单片机·mcu·学习·嵌入式·freertos·实时操作系统
w_zero_one8 小时前
算法--构造
java·数据结构·算法
热门推荐
01国内可直接用、免费额度/永久免费的大模型API清单（含 SiliconFlow、火山、阿里、智谱、百度、Kimi、DeepSeek、DMXAPI 等）02历年考研数学一、数学二、数学三真题试卷及答案PDF03GitHub 镜像站点04几个好用的ip纯净度检测网站05昇腾910B系列芯片完全指南：四款型号对比、服务器选型与大模型部署实战06苹果iPA游戏软件资源下载网站07蜘蛛磁力 搜索引擎大全，如何使用蜘蛛磁力查找磁力链接08如何新建文件夹？ 电脑新建文件夹的4种方法09Dify：一站式AI应用开发平台全解析与Windows部署指南10微信历史版本含下载地址（ Windows PC | 安卓 | MAC ）及设置微信不更新