2025.2.14电脑cpu不足以支撑训练，故尝试租用服务器
1.租用服务器
2.下载Xshell（未成功）
https://blog.youkuaiyun.com/m0_67400972/article/details/125346023
3.但是最后还是使用windows powershell进行连接的
4.然后跟着李沐
安装miniconda
https://blog.youkuaiyun.com/2301_76831056/article/details/143165738
安装速度太慢，使用镜像
pip install torch torchvision -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com5.将数据集传入jupyter
安装zip
将数据集解压
6.安装scikit-learn
pip install scikit-learn -i https://pypi.tuna.tsinghua.edu.cn/simple7.配置GPU
import torch
from torch import nn
torch.device('cpu'),torch.cuda.device('cuda'),torch.cuda.device('cuda:0')结果：
(device(type='cpu'), <torch.cuda.device at 0x7f77dd709d00>, <torch.cuda.device at 0x7f77dd709820>)
8.将LSTM 模型代码与交叉验证代码结合，搭建模型
import torch
import torch.nn as nn
from torch.utils.data import DataLoader, TensorDataset
import numpy as np
from sklearn.model_selection import KFold
# 1. 定义LSTM模型
class LSTMModel(nn.Module):
# ... (与您提供的代码相同)
def __init__(self, input_size, hidden_size, num_layers, num_classes):
super(LSTMModel, self).__init__()
self.hidden_size = hidden_size
self.num_layers = num_layers
# 定义LSTM层
self.lstm = nn.LSTM(
input_size=input_size,
hidden_size=hidden_size,
num_layers=num_layers,
batch_first=True # 输入输出张量格式为(batch, seq_len, feature)
)
# 定义全连接层
self.fc = nn.Linear(hidden_size, num_classes)
def forward(self, x):
# 初始化隐藏状态和细胞状态
h0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).to(x.device)
c0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).to(x.device)
# 前向传播LSTM
out, _ = self.lstm(x, (h0, c0)) # 输出维度(batch_size, seq_length, hidden_size)
# 只取最后一个时间步的输出
out = out[:, -1, :]
# 全连接层
out = self.fc(out)
return out
# 2. 超参数设置
input_size = 21
hidden_size = 128
num_layers = 2
num_classes = 2
batch_size = 64
learning_rate = 0.001
num_epochs = 10
k = 5 # 折数
# 3. 数据准备 (假设您的数据是numpy数组格式)
# X.shape = (31549, 600, 21)
# y.shape = (31549,)
# 4. KFold 交叉验证
kf = KFold(n_splits=k, shuffle=True, random_state=42)
for fold, (train_index, val_index) in enumerate(kf.split(X, y)):
print(f"Fold {fold + 1}:")
# 划分训练集和验证集
X_train, X_val = X[train_index], X[val_index]
y_train, y_val = y[train_index], y[val_index]
# 转换为PyTorch数据集
train_dataset = TensorDataset(
torch.from_numpy(X_train).float(),
torch.from_numpy(y_train).long()
)
val_dataset = TensorDataset(
torch.from_numpy(X_val).float(),
torch.from_numpy(y_val).long()
)
# 创建数据加载器
train_loader = DataLoader(
dataset=train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=2
)
val_loader = DataLoader(
dataset=val_dataset,
batch_size=batch_size,
shuffle=False, # 验证集不需要打乱
num_workers=2
)
# 5. 初始化模型
#device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = LSTMModel(input_size, hidden_size, num_layers, num_classes).to(device)
# 6. 定义损失函数和优化器
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
# 7. 训练循环
for epoch in range(num_epochs):
for i, (sequences, labels) in enumerate(train_loader):
sequences = sequences.to(device)
labels = labels.to(device)
# 前向传播
outputs = model(sequences)
loss = criterion(outputs, labels)
# 反向传播和优化
optimizer.zero_grad()
loss.backward()
optimizer.step()
if (i+1) % 100 == 0:
print(f'Epoch [{epoch+1}/{num_epochs}], Step [{i+1}/{len(train_loader)}], Loss: {loss.item():.4f}')
# 8. 验证
correct = 0
total = 0
with torch.no_grad():
for sequences, labels in val_loader:
sequences = sequences.to(device)
labels = labels.to(device)
outputs = model(sequences)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print(f"Epoch [{epoch+1}/{num_epochs}], Validation Accuracy: {100 * correct / total:.2f}%")
print("-" * 20)8.训练结果：
Fold 1: Epoch [1/10], Step [100/395], Loss: 0.5017 Epoch [1/10], Step [200/395], Loss: 0.5464 Epoch [1/10], Step [300/395], Loss: 0.7033 Epoch [1/10], Validation Accuracy: 77.08% Epoch [2/10], Step [100/395], Loss: 0.3463 Epoch [2/10], Step [200/395], Loss: 0.6087 Epoch [2/10], Step [300/395], Loss: 0.7428 Epoch [2/10], Validation Accuracy: 76.20% Epoch [3/10], Step [100/395], Loss: 0.5385 Epoch [3/10], Step [200/395], Loss: 0.3845 Epoch [3/10], Step [300/395], Loss: 0.4977 Epoch [3/10], Validation Accuracy: 79.67% Epoch [4/10], Step [100/395], Loss: 0.3336 Epoch [4/10], Step [200/395], Loss: 0.4343 Epoch [4/10], Step [300/395], Loss: 0.3964 Epoch [4/10], Validation Accuracy: 79.37% Epoch [5/10], Step [100/395], Loss: 0.4464 Epoch [5/10], Step [200/395], Loss: 0.3734 Epoch [5/10], Step [300/395], Loss: 0.2493 Epoch [5/10], Validation Accuracy: 90.54% Epoch [6/10], Step [100/395], Loss: 0.0937 Epoch [6/10], Step [200/395], Loss: 0.3071 Epoch [6/10], Step [300/395], Loss: 0.1935 Epoch [6/10], Validation Accuracy: 96.85% Epoch [7/10], Step [100/395], Loss: 0.1393 Epoch [7/10], Step [200/395], Loss: 0.1384 Epoch [7/10], Step [300/395], Loss: 0.0738 Epoch [7/10], Validation Accuracy: 97.50% Epoch [8/10], Step [100/395], Loss: 0.1021 Epoch [8/10], Step [200/395], Loss: 0.0307 Epoch [8/10], Step [300/395], Loss: 0.1038 Epoch [8/10], Validation Accuracy: 94.90% Epoch [9/10], Step [100/395], Loss: 0.1032 Epoch [9/10], Step [200/395], Loss: 0.0139 Epoch [9/10], Step [300/395], Loss: 0.0031 Epoch [9/10], Validation Accuracy: 97.75% Epoch [10/10], Step [100/395], Loss: 0.0714 Epoch [10/10], Step [200/395], Loss: 0.1020 Epoch [10/10], Step [300/395], Loss: 0.0334 Epoch [10/10], Validation Accuracy: 99.02% -------------------- Fold 2: Epoch [1/10], Step [100/395], Loss: 0.4848 Epoch [1/10], Step [200/395], Loss: 0.4125 Epoch [1/10], Step [300/395], Loss: 0.4249 Epoch [1/10], Validation Accuracy: 79.46% Epoch [2/10], Step [100/395], Loss: 0.6724 Epoch [2/10], Step [200/395], Loss: 0.4669 Epoch [2/10], Step [300/395], Loss: 0.5287 Epoch [2/10], Validation Accuracy: 78.13% Epoch [3/10], Step [100/395], Loss: 0.2836 Epoch [3/10], Step [200/395], Loss: 0.4645 Epoch [3/10], Step [300/395], Loss: 0.3780 Epoch [3/10], Validation Accuracy: 80.40% Epoch [4/10], Step [100/395], Loss: 0.2461 Epoch [4/10], Step [200/395], Loss: 0.3173 Epoch [4/10], Step [300/395], Loss: 0.1452 Epoch [4/10], Validation Accuracy: 90.98% Epoch [5/10], Step [100/395], Loss: 0.3913 Epoch [5/10], Step [200/395], Loss: 0.2494 Epoch [5/10], Step [300/395], Loss: 0.3230 Epoch [5/10], Validation Accuracy: 93.58% Epoch [6/10], Step [100/395], Loss: 0.0816 Epoch [6/10], Step [200/395], Loss: 0.1794 Epoch [6/10], Step [300/395], Loss: 0.2768 Epoch [6/10], Validation Accuracy: 92.49% Epoch [7/10], Step [100/395], Loss: 0.1752 Epoch [7/10], Step [200/395], Loss: 0.1088 Epoch [7/10], Step [300/395], Loss: 0.2150 Epoch [7/10], Validation Accuracy: 96.48% Epoch [8/10], Step [100/395], Loss: 0.1164 Epoch [8/10], Step [200/395], Loss: 0.1151 Epoch [8/10], Step [300/395], Loss: 0.1561 Epoch [8/10], Validation Accuracy: 96.01% Epoch [9/10], Step [100/395], Loss: 0.0974 Epoch [9/10], Step [200/395], Loss: 0.1123 Epoch [9/10], Step [300/395], Loss: 0.1261 Epoch [9/10], Validation Accuracy: 95.83% Epoch [10/10], Step [100/395], Loss: 0.0982 Epoch [10/10], Step [200/395], Loss: 0.0444 Epoch [10/10], Step [300/395], Loss: 0.0461 Epoch [10/10], Validation Accuracy: 97.89% -------------------- Fold 3: Epoch [1/10], Step [100/395], Loss: 0.4977 Epoch [1/10], Step [200/395], Loss: 0.5673 Epoch [1/10], Step [300/395], Loss: 0.4354 Epoch [1/10], Validation Accuracy: 84.33% Epoch [2/10], Step [100/395], Loss: 0.3990 Epoch [2/10], Step [200/395], Loss: 0.5802 Epoch [2/10], Step [300/395], Loss: 0.3587 Epoch [2/10], Validation Accuracy: 81.89% Epoch [3/10], Step [100/395], Loss: 0.4083 Epoch [3/10], Step [200/395], Loss: 0.4352 Epoch [3/10], Step [300/395], Loss: 0.3774 Epoch [3/10], Validation Accuracy: 69.29% Epoch [4/10], Step [100/395], Loss: 0.5158 Epoch [4/10], Step [200/395], Loss: 0.5685 Epoch [4/10], Step [300/395], Loss: 0.4904 Epoch [4/10], Validation Accuracy: 82.47% Epoch [5/10], Step [100/395], Loss: 0.3287 Epoch [5/10], Step [200/395], Loss: 0.2838 Epoch [5/10], Step [300/395], Loss: 0.2876 Epoch [5/10], Validation Accuracy: 85.36% Epoch [6/10], Step [100/395], Loss: 0.2592 Epoch [6/10], Step [200/395], Loss: 0.3600 Epoch [6/10], Step [300/395], Loss: 0.2365 Epoch [6/10], Validation Accuracy: 87.70% Epoch [7/10], Step [100/395], Loss: 0.3201 Epoch [7/10], Step [200/395], Loss: 0.3716 Epoch [7/10], Step [300/395], Loss: 0.2927 Epoch [7/10], Validation Accuracy: 86.96% Epoch [8/10], Step [100/395], Loss: 0.3415 Epoch [8/10], Step [200/395], Loss: 0.2263 Epoch [8/10], Step [300/395], Loss: 0.1731 Epoch [8/10], Validation Accuracy: 90.10% Epoch [9/10], Step [100/395], Loss: 0.2893 Epoch [9/10], Step [200/395], Loss: 0.3388 Epoch [9/10], Step [300/395], Loss: 0.3509 Epoch [9/10], Validation Accuracy: 89.52% Epoch [10/10], Step [100/395], Loss: 0.2174 Epoch [10/10], Step [200/395], Loss: 0.1554 Epoch [10/10], Step [300/395], Loss: 0.1092 Epoch [10/10], Validation Accuracy: 95.82% -------------------- Fold 4: Epoch [1/10], Step [100/395], Loss: 0.5843 Epoch [1/10], Step [200/395], Loss: 0.4975 Epoch [1/10], Step [300/395], Loss: 0.3452 Epoch [1/10], Validation Accuracy: 69.56% Epoch [2/10], Step [100/395], Loss: 0.4956 Epoch [2/10], Step [200/395], Loss: 0.4836 Epoch [2/10], Step [300/395], Loss: 0.4534 Epoch [2/10], Validation Accuracy: 83.30% Epoch [3/10], Step [100/395], Loss: 0.3182 Epoch [3/10], Step [200/395], Loss: 0.4520 Epoch [3/10], Step [300/395], Loss: 0.2990 Epoch [3/10], Validation Accuracy: 82.71% Epoch [4/10], Step [100/395], Loss: 0.4792 Epoch [4/10], Step [200/395], Loss: 0.3710 Epoch [4/10], Step [300/395], Loss: 0.5065 Epoch [4/10], Validation Accuracy: 81.43% Epoch [5/10], Step [100/395], Loss: 0.3549 Epoch [5/10], Step [200/395], Loss: 0.2655 Epoch [5/10], Step [300/395], Loss: 0.5377 Epoch [5/10], Validation Accuracy: 80.57% Epoch [6/10], Step [100/395], Loss: 0.7786 Epoch [6/10], Step [200/395], Loss: 0.4332 Epoch [6/10], Step [300/395], Loss: 0.4119 Epoch [6/10], Validation Accuracy: 75.50% Epoch [7/10], Step [100/395], Loss: 0.3652 Epoch [7/10], Step [200/395], Loss: 0.3263 Epoch [7/10], Step [300/395], Loss: 0.2774 Epoch [7/10], Validation Accuracy: 88.70% Epoch [8/10], Step [100/395], Loss: 0.2723 Epoch [8/10], Step [200/395], Loss: 0.2825 Epoch [8/10], Step [300/395], Loss: 0.3172 Epoch [8/10], Validation Accuracy: 88.83% Epoch [9/10], Step [100/395], Loss: 0.2246 Epoch [9/10], Step [200/395], Loss: 0.3250 Epoch [9/10], Step [300/395], Loss: 0.2093 Epoch [9/10], Validation Accuracy: 88.97% Epoch [10/10], Step [100/395], Loss: 0.3092 Epoch [10/10], Step [200/395], Loss: 0.1830 Epoch [10/10], Step [300/395], Loss: 0.2820 Epoch [10/10], Validation Accuracy: 92.54% -------------------- Fold 5: Epoch [1/10], Step [100/395], Loss: 0.3261 Epoch [1/10], Step [200/395], Loss: 0.5698 Epoch [1/10], Step [300/395], Loss: 0.4739 Epoch [1/10], Validation Accuracy: 78.92% Epoch [2/10], Step [100/395], Loss: 0.3079 Epoch [2/10], Step [200/395], Loss: 0.4412 Epoch [2/10], Step [300/395], Loss: 0.2897 Epoch [2/10], Validation Accuracy: 83.44% Epoch [3/10], Step [100/395], Loss: 0.4870 Epoch [3/10], Step [200/395], Loss: 0.2603 Epoch [3/10], Step [300/395], Loss: 0.4270 Epoch [3/10], Validation Accuracy: 89.92% Epoch [4/10], Step [100/395], Loss: 0.7423 Epoch [4/10], Step [200/395], Loss: 0.4999 Epoch [4/10], Step [300/395], Loss: 0.4962 Epoch [4/10], Validation Accuracy: 82.37% Epoch [5/10], Step [100/395], Loss: 0.6103 Epoch [5/10], Step [200/395], Loss: 0.4024 Epoch [5/10], Step [300/395], Loss: 0.4403 Epoch [5/10], Validation Accuracy: 84.83% Epoch [6/10], Step [100/395], Loss: 0.3716 Epoch [6/10], Step [200/395], Loss: 0.3998 Epoch [6/10], Step [300/395], Loss: 0.2721 Epoch [6/10], Validation Accuracy: 87.62% Epoch [7/10], Step [100/395], Loss: 0.4655 Epoch [7/10], Step [200/395], Loss: 0.4612 Epoch [7/10], Step [300/395], Loss: 0.4656 Epoch [7/10], Validation Accuracy: 74.27% Epoch [8/10], Step [100/395], Loss: 0.3697 Epoch [8/10], Step [200/395], Loss: 0.3434 Epoch [8/10], Step [300/395], Loss: 0.3524 Epoch [8/10], Validation Accuracy: 87.26% Epoch [9/10], Step [100/395], Loss: 0.4635 Epoch [9/10], Step [200/395], Loss: 0.1440 Epoch [9/10], Step [300/395], Loss: 0.1623 Epoch [9/10], Validation Accuracy: 93.36% Epoch [10/10], Step [100/395], Loss: 0.2480 Epoch [10/10], Step [200/395], Loss: 0.0923 Epoch [10/10], Step [300/395], Loss: 0.2037 Epoch [10/10], Validation Accuracy: 95.88% --------------------
结果解释：
这段输出是你的 LSTM 模型在进行 5 折交叉验证训练过程中的日志信息。它详细展示了每个 epoch（训练周期）和每个 step（步骤）的训练损失以及验证准确率。下面我将逐行解释这段输出的含义：
整体结构：
-
Fold 1, Fold 2, ..., Fold 5: 表示 5 折交叉验证的每一折。模型在每一折上都会进行训练和验证。
-
Epoch [X/10]: 表示当前是第几个 epoch，总共训练 10 个 epoch。一个 epoch 指的是模型完整地遍历一次训练数据集。
-
Step [X/395]: 表示当前是第几个 step，总共有 395 个 step。一个 step 通常指的是一个 batch 的训练过程。由于你的数据量和 batch size，计算得出每折有 395 个 step。
-
Loss: X.XXXX: 表示当前 step 的训练损失值。损失值越低，通常表示模型在当前 batch 上的表现越好。
-
Validation Accuracy: X.XX%: 表示当前 epoch 结束后，模型在验证集上的准确率。准确率越高，表示模型在验证集上的表现越好。
-
--------------------: 用于分隔不同折的训练结果。
具体解释（以 Fold 1 为例）：
-
Fold 1:: 表示当前是第一折交叉验证。 -
Epoch [1/10], Step [100/395], Loss: 0.5017: 表示在第一折的第一个 epoch 中，第 100 个 step 的训练损失值为 0.5017。 -
Epoch [1/10], Validation Accuracy: 77.08%: 表示在第一折的第一个 epoch 结束后，模型在验证集上的准确率为 77.08%。 -
Epoch [2/10], Step [200/395], Loss: 0.6087: 表示在第一折的第二个 epoch 中，第 200 个 step 的训练损失值为 0.6087。 -
Epoch [2/10], Validation Accuracy: 76.20%: 表示在第一折的第二个 epoch 结束后，模型在验证集上的准确率为 76.20%。 -
以此类推，直到 Fold 5 的训练结束。
分析这段输出：
-
训练损失 (Loss)： 训练损失在每个 epoch 中通常会逐渐下降，表示模型在不断学习和优化。
-
验证准确率 (Validation Accuracy)： 验证准确率是评估模型泛化能力的重要指标。在理想情况下，验证准确率应该随着 epoch 的增加而逐渐提高，但也有可能出现过拟合的情况，即模型在训练集上表现很好，但在验证集上表现不佳。
-
交叉验证： 通过 5 折交叉验证，你可以更全面地评估模型的性能，避免模型在单一数据集上的过拟合。
总结
这段输出详细记录了 LSTM 模型在 5 折交叉验证过程中的训练情况。通过分析损失值和验证准确率，你可以了解模型的训练效果，并根据需要调整模型参数或训练策略。如果你想得到最终的评估结果，通常需要计算 5 折验证准确率的平均值和标准差。
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
