- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍖 原作者：K同学啊
👉 学习目标：
● 思考本案例是否还有进一步优化的空间
🏡 我的环境：
● 语言环境：Python3.12.4
● 编译器：Jupyter Lab
● 深度学习框架：pyTorch
● 数据地址：上周的数据
一、数据预处理
1、设置GPU
import torch.nn as nn
import torch.nn.functional as F
import torchvision,torch
# 设置硬件设备，如果有GPU则使用，没有则使用cpu
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device
代码输出结果：
device(type='cuda')
2、数据导入
import numpy as np
import pandas as pd
import seaborn as sns
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
plt.rcParams['savefig.dpi'] = 500 # 图片像素
plt.rcParams['figure.dpi'] = 500 # 分辨率
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
import warnings
warnings.filterwarnings("ignore")
DataFrame = pd.read_excel('./data/dia.xls')
DataFrame.head()
代码输出结果：
卡号 性别 年龄 高密度脂蛋白胆固醇 低密度脂蛋白胆固醇 极低密度脂蛋白胆固醇 甘油三酯 总胆固醇 脉搏 舒张压 \
0 18054421 0 38 1.25 2.99 1.07 0.64 5.31 83 83
1 18054422 0 31 1.15 1.99 0.84 0.50 3.98 85 63
2 18054423 0 27 1.29 2.21 0.69 0.60 4.19 73 61
3 18054424 0 33 0.93 2.01 0.66 0.84 3.60 83 60
4 18054425 0 36 1.17 2.83 0.83 0.73 4.83 85 67
高血压史 尿素氮 尿酸 肌酐 体重检查结果 是否糖尿病
0 0 4.99 243.3 50 1 0
1 0 4.72 391.0 47 1 0
2 0 5.87 325.7 51 1 0
3 0 2.40 203.2 40 2 0
4 0 4.09 236.8 43 0 0
我们查看数据的shape：
DataFrame.shape
代码输出结果：
(1006, 16)
3、数据检查
# 查看数据是否有缺失值
print('数据缺失值---------------------------------')
print(DataFrame.isnull().sum())
代码输出结果：
数据缺失值---------------------------------
卡号 0
性别 0
年龄 0
高密度脂蛋白胆固醇 0
低密度脂蛋白胆固醇 0
极低密度脂蛋白胆固醇 0
甘油三酯 0
总胆固醇 0
脉搏 0
舒张压 0
高血压史 0
尿素氮 0
尿酸 0
肌酐 0
体重检查结果 0
是否糖尿病 0
dtype: int64
二、数据分析
1、数据分布分析
feature_map = {
'年龄': '年龄',
'高密度脂蛋白胆固醇': '高密度脂蛋白胆固醇',
'低密度脂蛋白胆固醇': '低密度脂蛋白胆固醇',
'极低密度脂蛋白胆固醇': '极低密度脂蛋白胆固醇',
'甘油三酯': '甘油三酯',
'总胆固醇': '总胆固醇',
'脉搏': '脉搏',
'舒张压': '舒张压',
'高血压史': '高血压史',
'尿素氮': '尿素氮',
'尿酸': '尿酸',
'肌酐': '肌酐',
'体重检查结果': '体重检查结果'
}
plt.figure(figsize=(15, 10))
for i, (col, col_name) in enumerate(feature_map.items(), 1):
plt.subplot(3, 5, i)
sns.boxplot(x='是否糖尿病', y=col, data=DataFrame, palette='Set2') # 使用不同颜色
plt.title(f'{col_name}的箱线图', fontsize=14)
plt.ylabel('数值', fontsize=12)
plt.grid(axis='y', linestyle='--', alpha=0.7)
plt.tight_layout()
plt.show()
代码输出结果：
2、相关性分析
import plotly
import plotly.express as px
# 删除列 '卡号'
DataFrame.drop(columns=['卡号'], inplace=True)
# 计算各列之间的相关系数
df_corr = DataFrame.corr()
# 相关矩阵生成函数
def corr_generate(df):
fig = px.imshow(df, text_auto=True, aspect="auto", color_continuous_scale='RdBu_r')
fig.show()
# 生成相关矩阵
corr_generate(df_corr)
代码输出结果：
三、LSTM模型
1、划分数据集
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import MinMaxScaler
# '高密度脂蛋白胆固醇'字段与糖尿病负相关，故而在 X 中去掉该字段
X = DataFrame.drop(['是否糖尿病', '高密度脂蛋白胆固醇'], axis=1)
y = DataFrame['是否糖尿病']
# 数据集标准化处理
sc_X = StandardScaler()
X = sc_X.fit_transform(X)
X = torch.tensor(np.array(X), dtype=torch.float32)
y = torch.tensor(np.array(y), dtype=torch.int64)
train_X, test_X, train_y, test_y = train_test_split(X, y,
test_size=0.2,
random_state=1)
# 变成 LSTM 所需形状：(batch_size, seq_len=1, input_size=13)
train_X = train_X.unsqueeze(1) # [643, 1, 13]
test_X = test_X.unsqueeze(1) # [161, 1, 13]
train_X.shape, train_y.shape
代码输出结果：
(torch.Size([804, 1, 13]), torch.Size([804]))
2、数据集构建
from torch.utils.data import TensorDataset, DataLoader
train_dl = DataLoader(TensorDataset(train_X, train_y),
batch_size=64,
shuffle=True)
test_dl = DataLoader(TensorDataset(test_X, test_y),
batch_size=64,
shuffle=True)
3、定义模型
我们在原来K同学的模型基础上，加入Attention机制：
class Attention(nn.Module):
def __init__(self, hidden_dim):
super(Attention, self).__init__()
self.attn = nn.Linear(hidden_dim, 1) # 输出为1个attention score
def forward(self, lstm_output): # lstm_output: (batch, seq_len, hidden_dim)
attn_weights = self.attn(lstm_output) # (batch, seq_len, 1)
attn_weights = torch.softmax(attn_weights, dim=1) # 时间维度归一化
context = torch.sum(attn_weights * lstm_output, dim=1) # (batch, hidden_dim)
return context
class model_lstm(nn.Module):
def __init__(self):
super(model_lstm, self).__init__()
self.lstm0 = nn.LSTM(input_size=13, hidden_size=200, num_layers=1, batch_first=True)
self.lstm1 = nn.LSTM(input_size=200, hidden_size=200, num_layers=1, batch_first=True)
self.attn = Attention(hidden_dim=200) # 添加 Attention 层
self.fc0 = nn.Linear(200, 2) # 输出维度根据你的任务设置为2类分类
def forward(self, x):
out, hidden1 = self.lstm0(x)
out, _ = self.lstm1(out, hidden1)
attn_out = self.attn(out) # (batch, hidden_dim)
out = self.fc0(attn_out) # 只使用 Attention 后的上下文向量进行分类
return out
model = model_lstm().to(device)
model
代码输出结果：
model_lstm(
(lstm0): LSTM(13, 200, batch_first=True)
(lstm1): LSTM(200, 200, batch_first=True)
(attn): Attention(
(attn): Linear(in_features=200, out_features=1, bias=True)
)
(fc0): Linear(in_features=200, out_features=2, bias=True)
)
四、训练模型
# 训练循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset) # 训练集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
train_loss, train_acc = 0, 0 # 初始化训练损失和正确率
for X, y in dataloader: # 获取图片及其标签
X, y = X.to(device), y.to(device)
# 计算预测误差
pred = model(X) # 网络输出
loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
# 反向传播
optimizer.zero_grad() # grad属性归零
loss.backward() # 反向传播
optimizer.step() # 每一步自动更新
# 记录acc与loss
train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
train_loss += loss.item()
train_acc /= size
train_loss /= num_batches
return train_acc, train_loss
def test(dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
test_loss, test_acc = 0, 0
# 当不进行训练时，停止梯度更新，节省计算内存消耗
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
# 计算loss
target_pred = model(imgs)
loss = loss_fn(target_pred, target)
test_loss += loss.item()
test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()
test_acc /= size
test_loss /= num_batches
return test_acc, test_loss
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
learn_rate = 1e-3 # 学习率
opt = torch.optim.Adam(model.parameters(), lr=learn_rate)
epochs = 50
train_loss = []
train_acc = []
test_loss = []
test_acc = []
for epoch in range(epochs):
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, opt)
model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
train_acc.append(epoch_train_acc)
train_loss.append(epoch_train_loss)
test_acc.append(epoch_test_acc)
test_loss.append(epoch_test_loss)
# 获取当前的学习率
lr = opt.state_dict()['param_groups'][0]['lr']
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
print(template.format(epoch + 1, epoch_train_acc * 100, epoch_train_loss,
epoch_test_acc * 100, epoch_test_loss, lr))
print("=" * 20, 'Done', "=" * 20)
代码输出结果：
Epoch: 1, Train_acc:66.8%, Train_loss:0.644, Test_acc:75.2%, Test_loss:0.589, Lr:1.00E-03
Epoch: 2, Train_acc:75.1%, Train_loss:0.517, Test_acc:75.7%, Test_loss:0.467, Lr:1.00E-03
Epoch: 3, Train_acc:77.4%, Train_loss:0.462, Test_acc:76.7%, Test_loss:0.482, Lr:1.00E-03
Epoch: 4, Train_acc:78.9%, Train_loss:0.432, Test_acc:78.7%, Test_loss:0.435, Lr:1.00E-03
Epoch: 5, Train_acc:79.1%, Train_loss:0.418, Test_acc:78.7%, Test_loss:0.443, Lr:1.00E-03
Epoch: 6, Train_acc:79.9%, Train_loss:0.405, Test_acc:80.2%, Test_loss:0.424, Lr:1.00E-03
Epoch: 7, Train_acc:79.6%, Train_loss:0.396, Test_acc:80.2%, Test_loss:0.383, Lr:1.00E-03
Epoch: 8, Train_acc:80.2%, Train_loss:0.389, Test_acc:78.2%, Test_loss:0.369, Lr:1.00E-03
Epoch: 9, Train_acc:80.8%, Train_loss:0.380, Test_acc:79.7%, Test_loss:0.383, Lr:1.00E-03
Epoch:10, Train_acc:81.0%, Train_loss:0.373, Test_acc:79.7%, Test_loss:0.425, Lr:1.00E-03
Epoch:11, Train_acc:81.3%, Train_loss:0.368, Test_acc:79.2%, Test_loss:0.360, Lr:1.00E-03
Epoch:12, Train_acc:82.6%, Train_loss:0.361, Test_acc:78.7%, Test_loss:0.363, Lr:1.00E-03
Epoch:13, Train_acc:81.5%, Train_loss:0.351, Test_acc:82.2%, Test_loss:0.369, Lr:1.00E-03
Epoch:14, Train_acc:82.2%, Train_loss:0.348, Test_acc:79.7%, Test_loss:0.373, Lr:1.00E-03
Epoch:15, Train_acc:82.1%, Train_loss:0.336, Test_acc:81.7%, Test_loss:0.398, Lr:1.00E-03
Epoch:16, Train_acc:82.8%, Train_loss:0.333, Test_acc:81.2%, Test_loss:0.323, Lr:1.00E-03
Epoch:17, Train_acc:83.0%, Train_loss:0.329, Test_acc:82.2%, Test_loss:0.350, Lr:1.00E-03
Epoch:18, Train_acc:83.0%, Train_loss:0.330, Test_acc:82.7%, Test_loss:0.362, Lr:1.00E-03
Epoch:19, Train_acc:84.6%, Train_loss:0.313, Test_acc:81.2%, Test_loss:0.404, Lr:1.00E-03
Epoch:20, Train_acc:85.3%, Train_loss:0.310, Test_acc:82.2%, Test_loss:0.401, Lr:1.00E-03
Epoch:21, Train_acc:84.6%, Train_loss:0.317, Test_acc:83.2%, Test_loss:0.399, Lr:1.00E-03
Epoch:22, Train_acc:85.3%, Train_loss:0.307, Test_acc:83.2%, Test_loss:0.403, Lr:1.00E-03
Epoch:23, Train_acc:86.4%, Train_loss:0.300, Test_acc:83.2%, Test_loss:0.396, Lr:1.00E-03
Epoch:24, Train_acc:85.8%, Train_loss:0.311, Test_acc:82.7%, Test_loss:0.481, Lr:1.00E-03
Epoch:25, Train_acc:86.4%, Train_loss:0.289, Test_acc:82.2%, Test_loss:0.419, Lr:1.00E-03
Epoch:26, Train_acc:85.9%, Train_loss:0.298, Test_acc:82.2%, Test_loss:0.374, Lr:1.00E-03
Epoch:27, Train_acc:86.2%, Train_loss:0.287, Test_acc:80.7%, Test_loss:0.429, Lr:1.00E-03
Epoch:28, Train_acc:87.1%, Train_loss:0.293, Test_acc:82.2%, Test_loss:0.470, Lr:1.00E-03
Epoch:29, Train_acc:87.4%, Train_loss:0.289, Test_acc:80.7%, Test_loss:0.470, Lr:1.00E-03
Epoch:30, Train_acc:87.1%, Train_loss:0.290, Test_acc:81.7%, Test_loss:0.421, Lr:1.00E-03
Epoch:31, Train_acc:87.3%, Train_loss:0.284, Test_acc:80.2%, Test_loss:0.470, Lr:1.00E-03
Epoch:32, Train_acc:87.6%, Train_loss:0.279, Test_acc:80.2%, Test_loss:0.429, Lr:1.00E-03
Epoch:33, Train_acc:86.7%, Train_loss:0.290, Test_acc:79.7%, Test_loss:0.385, Lr:1.00E-03
Epoch:34, Train_acc:88.2%, Train_loss:0.273, Test_acc:80.2%, Test_loss:0.445, Lr:1.00E-03
Epoch:35, Train_acc:87.9%, Train_loss:0.277, Test_acc:78.2%, Test_loss:0.421, Lr:1.00E-03
Epoch:36, Train_acc:88.3%, Train_loss:0.268, Test_acc:79.2%, Test_loss:0.455, Lr:1.00E-03
Epoch:37, Train_acc:87.7%, Train_loss:0.264, Test_acc:79.7%, Test_loss:0.470, Lr:1.00E-03
Epoch:38, Train_acc:88.7%, Train_loss:0.267, Test_acc:80.2%, Test_loss:0.465, Lr:1.00E-03
Epoch:39, Train_acc:88.4%, Train_loss:0.266, Test_acc:78.2%, Test_loss:0.456, Lr:1.00E-03
Epoch:40, Train_acc:88.7%, Train_loss:0.261, Test_acc:78.7%, Test_loss:0.412, Lr:1.00E-03
Epoch:41, Train_acc:87.8%, Train_loss:0.265, Test_acc:79.2%, Test_loss:0.671, Lr:1.00E-03
Epoch:42, Train_acc:89.1%, Train_loss:0.254, Test_acc:78.7%, Test_loss:0.381, Lr:1.00E-03
Epoch:43, Train_acc:88.7%, Train_loss:0.254, Test_acc:77.7%, Test_loss:0.401, Lr:1.00E-03
Epoch:44, Train_acc:88.7%, Train_loss:0.246, Test_acc:78.2%, Test_loss:0.670, Lr:1.00E-03
Epoch:45, Train_acc:89.6%, Train_loss:0.246, Test_acc:78.7%, Test_loss:0.415, Lr:1.00E-03
Epoch:46, Train_acc:89.3%, Train_loss:0.244, Test_acc:78.7%, Test_loss:0.452, Lr:1.00E-03
Epoch:47, Train_acc:89.1%, Train_loss:0.246, Test_acc:78.2%, Test_loss:0.488, Lr:1.00E-03
Epoch:48, Train_acc:89.3%, Train_loss:0.245, Test_acc:79.7%, Test_loss:0.397, Lr:1.00E-03
Epoch:49, Train_acc:89.4%, Train_loss:0.237, Test_acc:78.2%, Test_loss:0.412, Lr:1.00E-03
Epoch:50, Train_acc:89.4%, Train_loss:0.237, Test_acc:78.7%, Test_loss:0.451, Lr:1.00E-03
==================== Done ====================
五、模型评估
import matplotlib.pyplot as plt
# 隐藏警告
import warnings
warnings.filterwarnings("ignore") # 忽略警告信息
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
plt.rcParams['figure.dpi'] = 100 # 分辨率
from datetime import datetime
current_time = datetime.now() # 获取当前时间
epochs_range = range(epochs)
plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.xlabel(current_time) # 打卡请带上时间戳，否则代码截图无效
plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
代码输出结果：
六、总结
加入attention机制后，训练集的准确度能够达到80%，但是出现了过拟合的状态。
扫一扫
698
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
