|
楼主 |
发表于 2024-10-22 22:21:03
|
显示全部楼层
#dnn模型(train_test_split)
import torch
import torch.nn as nn
from torch.utils import data
from torch.utils.data import Dataset,DataLoader
from torch import optim
#定义神经网络模型
dropout1, dropout2 = 0.3, 0.6
class SimpleNN(nn.Module):
def __init__(self):
super(SimpleNN, self).__init__() # 继承需要用 SimpleNN
self.dense = nn.Sequential(
nn.Flatten(),
nn.Linear(12, 128),
nn.ReLU(),
nn.Dropout(dropout1),
nn.Linear(128, 256),
nn.ReLU(),
nn.Dropout(dropout2),
nn.Linear(256, 1),
)
def forward(self, X):
x = self.dense(X)
output = torch.sigmoid(x)
return output
#初始化模型和优化器
simple_dnn_model = SimpleNN()
loss = nn.BCELoss() #定义损失函数
optimizer = optim.Adam(nn_model.parameters(),lr=0.0001) #定义优化器
#初始化列表
acc_list = []
loss_list = []
# 开始训练
batch_size = 99
num_epochs = 1000
#创建数据集
train_dataset = data.TensorDataset(X_train, y_train)
valid_dataset = data.TensorDataset(X_valid, y_valid)
# 获取一个数据迭代器
train_iter = DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=True,num_workers=2)#shuffle=True相当于sampler=RandomSampler(dataset)
valid_iter = DataLoader(dataset=valid_dataset,batch_size=batch_size,shuffle=True,num_workers=2)
#开始迭代
for epoch in range(num_epochs):
train_loss = 0
num_right = 0
for tensor_x, tensor_y in train_iter:#训练集执行梯度更新
tensor_x = tensor_x.float()
tensor_y = tensor_y.float().reshape(-1, 1)
optimizer.zero_grad() #梯度清零
pre_train = simple_dnn_model(tensor_x)
train_l = loss(pre_train, tensor_y) #损失应避免与全局变量loss重名
train_l.backward()#前向传播
optimizer.step()#梯度下降
train_loss += train_l.item() * len(tensor_x)#批量损失
result = [1 if out >= 0.5 else 0 for out in pre_train]
num_right += np.sum(np.array(result) == tensor_y.numpy().reshape(-1))
train_loss = train_loss / len(train_iter.dataset)
train_accuracy = num_right / len(train_iter.dataset)
if epoch % 200 == 0:
print('Loss: {} Accuracy: {} Epoch:{}'.format(train_loss, train_accuracy, epoch))
with torch.no_grad():
valid_loss = 0
num_right = 0
for tensor_x, tensor_y in valid_iter:
tensor_x = tensor_x.float()
tensor_y = tensor_y.float().reshape(-1, 1)
pre_valid = simple_dnn_model(tensor_x)
valid_l = loss(pre_valid, tensor_y)
valid_loss += valid_l.item() * len(tensor_x)
result = [1 if out >= 0.5 else 0 for out in pre_valid]
num_right += np.sum(np.array(result) == tensor_y.numpy().reshape(-1))
valid_loss = valid_loss / len(valid_iter.dataset)
valid_accuracy = num_right / len(valid_iter.dataset)
if epoch % 200 == 0:
print('Valid Loss: {} Accuracy: {} Epoch:{}'.format(valid_loss, valid_accuracy, epoch))
#将每次迭代的结果写入列表
loss_list.append(valid_loss)
acc_list.append(valid_accuracy)
print('Training Ended')
print('Average Loss: {} Average Accuracy: {}'.format(np.mean(loss_list), np.mean(acc_list)))
数据的特征和标签类型是tensor,转换为numpy后形状分别是X_train.values.shape,y_train.values.shape,X_valid.values.shape,y_valid.values.shape
((569, 12), (569,), (143, 12), (143,))
训练模型代码报错如下---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[63], line 47
43 num_epochs = 1000
46 #创建数据集
---> 47 train_dataset = data.TensorDataset(X_train, y_train)
48 valid_dataset = data.TensorDataset(X_valid, y_valid)
50 # 获取一个数据迭代器
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/dataset.py:204, in TensorDataset.__init__(self, *tensors)
203 def __init__(self, *tensors: Tensor) -> None:
--> 204 assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors), "Size mismatch between tensors"
205 self.tensors = tensors
File /opt/conda/lib/python3.10/site-packages/torch/utils/data/dataset.py:204, in <genexpr>(.0)
203 def __init__(self, *tensors: Tensor) -> None:
--> 204 assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors), "Size mismatch between tensors"
205 self.tensors = tensors
TypeError: 'int' object is not callable
|
|