深度学习训练camp-第R3周：RNN-心脏病预测

• 🍨 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

🍺要求：

1. 本地读取并加载数据。
2. 了解循环神经网络（RNN）的构建过程
3. 测试集accuracy到达87%

🍻拔高：

1. 测试集accuracy到达89%

🏡 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：🔗百度网盘

一、RNN简介

传统神经网络的结构比较简单：输入层-隐藏层-输出层

RNN和传统神经网络最大的区别在于每次都会将前一次的输出结果带入到下一次的隐藏层中，一起训练。

循环神经网络（RNN）是一种专门用于处理序列数据（sequence data）的神经网络架构。它的核心特点是具有循环连接，能够利用前面时刻的信息来影响当前时刻的计算。这使得 RNN 特别适用于时间序列数据、自然语言处理（NLP）和序列预测任务。

二、心脏病的预测

一、前期准备

import numpy as np
import pandas as pd
import torch
from torch import nn
import torch.nn.functional as F
import seaborn as sns

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device

输出：

device(type='cuda')

我的GPU是Nvdia RTX 4070Ti super

二、导入数据

🥂 数据介绍：

● age：1) 年龄

● sex：2) 性别

● cp：3) 胸痛类型 (4 values)

● trestbps：4) 静息血压

● chol：5) 血清胆甾醇 (mg/dl

● fbs：6) 空腹血糖 > 120 mg/dl

● restecg：7) 静息心电图结果 (值 0,1 ,2)

● thalach：8) 达到的最大心率

● exang：9) 运动诱发的心绞痛

● oldpeak：10) 相对于静止状态，运动引起的ST段压低

● slope：11) 运动峰值 ST 段的斜率

● ca：12) 荧光透视着色的主要血管数量 (0-3)

● thal：13) 0 = 正常；1 = 固定缺陷；2 = 可逆转的缺陷

● target：14) 0 = 心脏病发作的几率较小 1 = 心脏病发作的几率更大

源数据属于K同学🍖 原作者：K同学啊

file_path = 'D:/OneDrive/code learning(python and matlab and latex)/365camp/data/heart.csv'
df = pd.read_csv(file_path)
df

代码输出：

     age  sex  cp  trestbps  chol  fbs  restecg  thalach  exang  oldpeak  \
0     63    1   3       145   233    1        0      150      0      2.3   
1     37    1   2       130   250    0        1      187      0      3.5   
2     41    0   1       130   204    0        0      172      0      1.4   
3     56    1   1       120   236    0        1      178      0      0.8   
4     57    0   0       120   354    0        1      163      1      0.6   
..   ...  ...  ..       ...   ...  ...      ...      ...    ...      ...   
298   57    0   0       140   241    0        1      123      1      0.2   
299   45    1   3       110   264    0        1      132      0      1.2   
300   68    1   0       144   193    1        1      141      0      3.4   
301   57    1   0       130   131    0        1      115      1      1.2   
302   57    0   1       130   236    0        0      174      0      0.0   

     slope  ca  thal  target  
0        0   0     1       1  
1        0   0     2       1  
2        2   0     2       1  
3        2   0     2       1  
4        2   0     2       1  
..     ...  ..   ...     ...  
298      1   0     3       0  
299      1   0     3       0  
300      1   2     3       0  
301      1   1     3       0  
302      1   1     2       0  

[303 rows x 14 columns]

三、数据预处理

1、划分训练集和测试集

from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split

x = df.iloc[:, :-1]
y = df.iloc[:,-1]

sc = StandardScaler()
X = sc.fit_transform(x)

X = torch.tensor(np.array(X), dtype=torch.float32)
y = torch.tensor(np.array(y), dtype=torch.int64)

X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    test_size=0.1,
                                                    random_state=1
                                                    )

X_train.shape, y_train.shape

代码输出：

(torch.Size([272, 13]), torch.Size([272]))

2、标准化

from torch.utils.data import TensorDataset, DataLoader, dataloader

train_dl = DataLoader(TensorDataset(X_train, y_train),
                      batch_size=64,
                      shuffle=False)

test_dl = DataLoader(TensorDataset(X_test, y_test),
                     batch_size=64,
                     shuffle=False)

输出：

(<torch.utils.data.dataloader.DataLoader at 0x21cbfd08b90>,
 <torch.utils.data.dataloader.DataLoader at 0x21cbfd0adb0>)

四、构建RNN模型

class model_RNN(nn.Module):
    def __init__(self):
        super(model_RNN, self).__init__()
        self.rnn0 = nn.RNN(input_size=13, hidden_size=200, num_layers=1, batch_first=True)
        
        self.fc0 = nn.Linear(in_features=200, out_features=50)
        self.fc1 = nn.Linear(in_features=50,out_features=2)
        
    def forward(self, x):
        out, hidden1 = self.rnn0(x)
        out = self.fc0(out)
        out = self.fc1(out)
        return out
    
model = model_RNN().to(device)
model

代码输出：

model_RNN(
  (rnn0): RNN(13, 200, batch_first=True)
  (fc0): Linear(in_features=200, out_features=50, bias=True)
  (fc1): Linear(in_features=50, out_features=2, bias=True)
)

五、编写训练函数和测试函数

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

六、正式训练

loss_fn = nn.CrossEntropyLoss()
learning_rate = 1e-4
opt = torch.optim.Adam(model.parameters(), lr=learning_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_loss.append(epoch_train_loss)
    train_acc.append(epoch_train_acc)
    test_loss.append(epoch_test_loss)
    test_acc.append(epoch_test_acc)
    
    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：

Epoch: 1, Train_acc:47.1%, Train_loss:0.689, Test_acc:67.7%, Test_loss:0.639, Lr:1.00E-04
Epoch: 2, Train_acc:54.0%, Train_loss:0.672, Test_acc:83.9%, Test_loss:0.622, Lr:1.00E-04
Epoch: 3, Train_acc:66.5%, Train_loss:0.653, Test_acc:80.6%, Test_loss:0.606, Lr:1.00E-04
Epoch: 4, Train_acc:76.5%, Train_loss:0.638, Test_acc:80.6%, Test_loss:0.590, Lr:1.00E-04
Epoch: 5, Train_acc:79.4%, Train_loss:0.621, Test_acc:77.4%, Test_loss:0.574, Lr:1.00E-04
Epoch: 6, Train_acc:80.1%, Train_loss:0.620, Test_acc:80.6%, Test_loss:0.558, Lr:1.00E-04
Epoch: 7, Train_acc:79.0%, Train_loss:0.604, Test_acc:77.4%, Test_loss:0.543, Lr:1.00E-04
Epoch: 8, Train_acc:82.0%, Train_loss:0.603, Test_acc:83.9%, Test_loss:0.529, Lr:1.00E-04
Epoch: 9, Train_acc:79.8%, Train_loss:0.585, Test_acc:83.9%, Test_loss:0.516, Lr:1.00E-04
Epoch:10, Train_acc:80.1%, Train_loss:0.564, Test_acc:83.9%, Test_loss:0.502, Lr:1.00E-04
Epoch:11, Train_acc:77.9%, Train_loss:0.559, Test_acc:83.9%, Test_loss:0.488, Lr:1.00E-04
Epoch:12, Train_acc:81.2%, Train_loss:0.542, Test_acc:83.9%, Test_loss:0.473, Lr:1.00E-04
Epoch:13, Train_acc:81.2%, Train_loss:0.543, Test_acc:87.1%, Test_loss:0.459, Lr:1.00E-04
Epoch:14, Train_acc:80.9%, Train_loss:0.529, Test_acc:83.9%, Test_loss:0.447, Lr:1.00E-04
Epoch:15, Train_acc:81.6%, Train_loss:0.520, Test_acc:83.9%, Test_loss:0.435, Lr:1.00E-04
Epoch:16, Train_acc:80.1%, Train_loss:0.506, Test_acc:83.9%, Test_loss:0.423, Lr:1.00E-04
Epoch:17, Train_acc:81.6%, Train_loss:0.507, Test_acc:83.9%, Test_loss:0.412, Lr:1.00E-04
Epoch:18, Train_acc:82.0%, Train_loss:0.472, Test_acc:83.9%, Test_loss:0.401, Lr:1.00E-04
Epoch:19, Train_acc:82.0%, Train_loss:0.478, Test_acc:83.9%, Test_loss:0.393, Lr:1.00E-04
Epoch:20, Train_acc:82.4%, Train_loss:0.485, Test_acc:83.9%, Test_loss:0.385, Lr:1.00E-04
Epoch:21, Train_acc:81.2%, Train_loss:0.459, Test_acc:83.9%, Test_loss:0.378, Lr:1.00E-04
Epoch:22, Train_acc:84.2%, Train_loss:0.465, Test_acc:83.9%, Test_loss:0.370, Lr:1.00E-04
Epoch:23, Train_acc:82.0%, Train_loss:0.442, Test_acc:83.9%, Test_loss:0.363, Lr:1.00E-04
Epoch:24, Train_acc:81.6%, Train_loss:0.433, Test_acc:83.9%, Test_loss:0.357, Lr:1.00E-04
Epoch:25, Train_acc:82.0%, Train_loss:0.425, Test_acc:87.1%, Test_loss:0.351, Lr:1.00E-04
Epoch:26, Train_acc:82.7%, Train_loss:0.423, Test_acc:87.1%, Test_loss:0.345, Lr:1.00E-04
Epoch:27, Train_acc:82.4%, Train_loss:0.428, Test_acc:87.1%, Test_loss:0.339, Lr:1.00E-04
Epoch:28, Train_acc:83.1%, Train_loss:0.443, Test_acc:87.1%, Test_loss:0.334, Lr:1.00E-04
Epoch:29, Train_acc:83.1%, Train_loss:0.433, Test_acc:87.1%, Test_loss:0.330, Lr:1.00E-04
Epoch:30, Train_acc:83.1%, Train_loss:0.407, Test_acc:87.1%, Test_loss:0.326, Lr:1.00E-04
Epoch:31, Train_acc:83.5%, Train_loss:0.415, Test_acc:87.1%, Test_loss:0.321, Lr:1.00E-04
Epoch:32, Train_acc:83.5%, Train_loss:0.389, Test_acc:87.1%, Test_loss:0.316, Lr:1.00E-04
Epoch:33, Train_acc:82.4%, Train_loss:0.400, Test_acc:87.1%, Test_loss:0.312, Lr:1.00E-04
Epoch:34, Train_acc:85.3%, Train_loss:0.380, Test_acc:87.1%, Test_loss:0.311, Lr:1.00E-04
Epoch:35, Train_acc:83.8%, Train_loss:0.388, Test_acc:87.1%, Test_loss:0.310, Lr:1.00E-04
Epoch:36, Train_acc:83.1%, Train_loss:0.377, Test_acc:87.1%, Test_loss:0.311, Lr:1.00E-04
Epoch:37, Train_acc:83.8%, Train_loss:0.382, Test_acc:87.1%, Test_loss:0.310, Lr:1.00E-04
Epoch:38, Train_acc:83.8%, Train_loss:0.379, Test_acc:87.1%, Test_loss:0.309, Lr:1.00E-04
Epoch:39, Train_acc:84.6%, Train_loss:0.388, Test_acc:87.1%, Test_loss:0.307, Lr:1.00E-04
Epoch:40, Train_acc:84.2%, Train_loss:0.373, Test_acc:87.1%, Test_loss:0.306, Lr:1.00E-04
Epoch:41, Train_acc:83.8%, Train_loss:0.373, Test_acc:87.1%, Test_loss:0.307, Lr:1.00E-04
Epoch:42, Train_acc:84.2%, Train_loss:0.371, Test_acc:87.1%, Test_loss:0.309, Lr:1.00E-04
Epoch:43, Train_acc:84.2%, Train_loss:0.363, Test_acc:87.1%, Test_loss:0.310, Lr:1.00E-04
Epoch:44, Train_acc:83.5%, Train_loss:0.361, Test_acc:87.1%, Test_loss:0.311, Lr:1.00E-04
Epoch:45, Train_acc:84.2%, Train_loss:0.394, Test_acc:87.1%, Test_loss:0.313, Lr:1.00E-04
Epoch:46, Train_acc:83.8%, Train_loss:0.380, Test_acc:87.1%, Test_loss:0.314, Lr:1.00E-04
Epoch:47, Train_acc:84.6%, Train_loss:0.365, Test_acc:87.1%, Test_loss:0.314, Lr:1.00E-04
Epoch:48, Train_acc:84.6%, Train_loss:0.338, Test_acc:87.1%, Test_loss:0.313, Lr:1.00E-04
Epoch:49, Train_acc:83.5%, Train_loss:0.373, Test_acc:87.1%, Test_loss:0.314, Lr:1.00E-04
Epoch:50, Train_acc:84.6%, Train_loss:0.355, Test_acc:87.1%, Test_loss:0.314, Lr:1.00E-04
==================== Done ====================

七、数据可视化

1、Acc以及Loss图

import matplotlib.pyplot as plt
from datetime import datetime
#隐藏警告
import warnings
warnings.filterwarnings("ignore")        #忽略警告信息

current_time = datetime.now() # 获取当前时间

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']         = 200        #分辨率

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()

输出：

2、混淆矩阵

import numpy as np
import matplotlib.pyplot as plt
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay

# 计算混淆矩阵
cm = confusion_matrix(y_test, pred)

plt.figure(figsize=(6,5))
plt.suptitle('')
sns.heatmap(cm, annot=True, fmt="d", cmap="Blues")

# 修改字体大小
plt.xticks(fontsize=10)
plt.yticks(fontsize=10)
plt.title("Confusion Matrix", fontsize=12)
plt.xlabel("Predicted Label", fontsize=10)
plt.ylabel("True Label", fontsize=10)

# 显示图
plt.tight_layout()  # 调整布局防止重叠
plt.show()

代码输出：

八、总结与提升

RNN可以简单理解为读书，在上一页的基础上对下一页进行理解

1. RNN 的 Forward 过程：模拟人类阅读故事
   假设你有一个句子：

"I love deep learning and PyTorch is amazing!"

RNN 的工作方式类似于逐字阅读：

读到 "I" → 形成初步理解。

读到 "love" → 结合 "I" 的信息，加深理解。

读到 "deep" → 结合前面所有的信息，调整理解。

读完整个句子 → 用最终的理解去做决策（比如预测句子的情感）。

这个过程就类似于 RNN 的 前向传播（forward pass）。

我们发现训练epoch次数上升后反而loss增加，因此我们将数据预处理的shuffle设置为true以及减少训练次数，可以得到一下的训练结果：

Epoch: 1, Train_acc:51.8%, Train_loss:0.696, Test_acc:54.8%, Test_loss:0.671, Lr:1.00E-04
Epoch: 2, Train_acc:64.3%, Train_loss:0.679, Test_acc:77.4%, Test_loss:0.653, Lr:1.00E-04
Epoch: 3, Train_acc:75.0%, Train_loss:0.660, Test_acc:87.1%, Test_loss:0.622, Lr:1.00E-04
Epoch: 4, Train_acc:77.2%, Train_loss:0.644, Test_acc:90.3%, Test_loss:0.599, Lr:1.00E-04
Epoch: 5, Train_acc:79.4%, Train_loss:0.625, Test_acc:87.1%, Test_loss:0.578, Lr:1.00E-04
Epoch: 6, Train_acc:78.7%, Train_loss:0.611, Test_acc:87.1%, Test_loss:0.566, Lr:1.00E-04
Epoch: 7, Train_acc:81.6%, Train_loss:0.595, Test_acc:93.5%, Test_loss:0.521, Lr:1.00E-04
Epoch: 8, Train_acc:77.9%, Train_loss:0.580, Test_acc:90.3%, Test_loss:0.520, Lr:1.00E-04
Epoch: 9, Train_acc:79.0%, Train_loss:0.565, Test_acc:90.3%, Test_loss:0.496, Lr:1.00E-04
Epoch:10, Train_acc:80.1%, Train_loss:0.545, Test_acc:90.3%, Test_loss:0.455, Lr:1.00E-04
Epoch:11, Train_acc:82.7%, Train_loss:0.528, Test_acc:87.1%, Test_loss:0.444, Lr:1.00E-04
Epoch:12, Train_acc:81.2%, Train_loss:0.512, Test_acc:93.5%, Test_loss:0.422, Lr:1.00E-04
Epoch:13, Train_acc:82.0%, Train_loss:0.493, Test_acc:90.3%, Test_loss:0.405, Lr:1.00E-04
Epoch:14, Train_acc:82.4%, Train_loss:0.477, Test_acc:90.3%, Test_loss:0.381, Lr:1.00E-04
Epoch:15, Train_acc:82.7%, Train_loss:0.469, Test_acc:87.1%, Test_loss:0.365, Lr:1.00E-04
Epoch:16, Train_acc:83.1%, Train_loss:0.457, Test_acc:90.3%, Test_loss:0.356, Lr:1.00E-04
Epoch:17, Train_acc:84.2%, Train_loss:0.431, Test_acc:87.1%, Test_loss:0.342, Lr:1.00E-04
Epoch:18, Train_acc:83.5%, Train_loss:0.435, Test_acc:87.1%, Test_loss:0.340, Lr:1.00E-04
Epoch:19, Train_acc:83.8%, Train_loss:0.424, Test_acc:90.3%, Test_loss:0.324, Lr:1.00E-04
Epoch:20, Train_acc:82.4%, Train_loss:0.416, Test_acc:90.3%, Test_loss:0.321, Lr:1.00E-04
Epoch:21, Train_acc:85.3%, Train_loss:0.422, Test_acc:90.3%, Test_loss:0.310, Lr:1.00E-04
Epoch:22, Train_acc:84.2%, Train_loss:0.408, Test_acc:90.3%, Test_loss:0.294, Lr:1.00E-04
Epoch:23, Train_acc:86.0%, Train_loss:0.388, Test_acc:87.1%, Test_loss:0.307, Lr:1.00E-04
Epoch:24, Train_acc:84.6%, Train_loss:0.385, Test_acc:90.3%, Test_loss:0.290, Lr:1.00E-04
Epoch:25, Train_acc:84.9%, Train_loss:0.381, Test_acc:87.1%, Test_loss:0.311, Lr:1.00E-04
Epoch:26, Train_acc:85.3%, Train_loss:0.368, Test_acc:87.1%, Test_loss:0.293, Lr:1.00E-04
Epoch:27, Train_acc:84.2%, Train_loss:0.388, Test_acc:87.1%, Test_loss:0.315, Lr:1.00E-04
Epoch:28, Train_acc:84.6%, Train_loss:0.376, Test_acc:87.1%, Test_loss:0.285, Lr:1.00E-04
Epoch:29, Train_acc:83.1%, Train_loss:0.375, Test_acc:87.1%, Test_loss:0.306, Lr:1.00E-04
Epoch:30, Train_acc:84.2%, Train_loss:0.379, Test_acc:87.1%, Test_loss:0.278, Lr:1.00E-04
Epoch:31, Train_acc:84.6%, Train_loss:0.374, Test_acc:87.1%, Test_loss:0.308, Lr:1.00E-04
Epoch:32, Train_acc:84.6%, Train_loss:0.368, Test_acc:90.3%, Test_loss:0.289, Lr:1.00E-04
==================== Done ====================

测试集的准确度超过了90%