逻辑回归要求:
完整代码为:
import matplotlib.pyplot as plt
import torch
import matplotlib.pyplot as pl
import numpy as npx_data = torch.Tensor([[1.0], [2.0], [3.0]])
y_data = torch.Tensor([[0], [0], [1]])
class LogisticRegressionModel(torch.nn.Module):
# 构造函数:初始化对象默认调用的函数
def __init__(self):
# 必写
super(LogisticRegressionModel, self).__init__()
# 构造对象
self.linear = torch.nn.Linear(1, 1)# 前馈任务所要进行的计算
def forward(self, x):
y_pred = torch.sigmoid(self.linear(x))
return y_pred
# 实例化
model = LogisticRegressionModel()
# 损失函数对象
criterion = torch.nn.BCELoss(reduction='sum')
# 优化器对象
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
# 进行训练
for epoch in range(1000):
# 计算y hat
y_pred = model(x_data)
# 计算损失
loss = criterion(y_pred, y_data)
print(epoch, loss.item())# 所有权重每次都梯度清零
optimizer.zero_grad()
# 反向传播求梯度
loss.backward()
# 更新,step()更新函数
optimizer.step()
# 画图
x = np.linspace(0, 10, 200)
x_t = torch.Tensor(x).view((200, 1))
y_t = model(x_t)
y = y_t.data.numpy()
plt.plot(x, y)
plt.plot([0, 10], [0.5, 0.5], c='r')
plt.xlabel('Hours')
plt.ylabel('Probability of Pass')
plt.show()
总结:
1.torch.Tensor和torch.tensor初始化的变量类型不一致会报错
2.pytorch的版本不一样,一些语句的修改
criterion = torch.nn.BCELoss(size_average=True)
改为:
criterion = torch.nn.BCELoss(reduction='mean')
criterion = torch.nn.BCELoss(size_average=False)
改为:
criterion = torch.nn.BCELoss(reduction='sum')