PyTorch示例——LogisticRegressionModel

版本信息

• PyTorch: 1.12.1
• Python: 3.7.13

导包

import torch
import matplotlib.pyplot as plt

原始数据

X_data = torch.Tensor([[1.0, 2.9], [2.0, 6.1], [3.0, 9.2], [4.0, 12.3], [5.0, 14.9], [6.0, 18.1]])
y_data = torch.Tensor([[0], [0], [0], [1], [1], [1]])

构建模型

# 构建模型
class MyLogisticRegressionModel(torch.nn.Module):
    
    def __init__(self):
        super(MyLogisticRegressionModel, self).__init__()
        self.linear = torch.nn.Linear(2, 1)
        self.logistic = torch.nn.Sigmoid()
        
    def forward(self, x):
        ouput = self.linear(x)
        ouput = self.logistic(ouput)
        return ouput

开始训练

• 代码

# 参数
learning_rate = 0.005
epoch_num = 100

# 构建模型
my_model = MyLogisticRegressionModel()
# 计算交叉熵损失
loss = torch.nn.BCELoss(reduction='sum')
# 优化器，即学习器，使用SGD，会更新模型参数my_model.parameters()|
optimizer = torch.optim.SGD(my_model.parameters(), lr=learning_rate)

# 开始训练
loss_list = []
for epoch in range(epoch_num):
    y_pred = my_model(X_data)
    l = loss(y_pred, y_data)
    
    optimizer.zero_grad() # 梯度置零
    l.backward() # 反向传播梯度
    optimizer.step() # 更新梯度
    
    print(f"Train... ===> epoch = {epoch}, loss_val = {l.item()}")
    loss_list.append(l.item())

• 输出信息

Train... ===> epoch = 0, loss = 6.798981189727783
Train... ===> epoch = 1, loss = 4.380733966827393
Train... ===> epoch = 2, loss = 4.372033596038818
Train... ===> epoch = 3, loss = 4.36480188369751
Train... ===> epoch = 4, loss = 4.357597351074219
Train... ===> epoch = 5, loss = 4.350410461425781
Train... ===> epoch = 6, loss = 4.343241214752197
Train... ===> epoch = 7, loss = 4.336090087890625
Train... ===> epoch = 8, loss = 4.328956604003906
Train... ===> epoch = 9, loss = 4.321841239929199
......
Train... ===> epoch = 90, loss = 3.8016786575317383
Train... ===> epoch = 91, loss = 3.7959144115448
Train... ===> epoch = 92, loss = 3.7901651859283447
Train... ===> epoch = 93, loss = 3.7844314575195312
Train... ===> epoch = 94, loss = 3.778712272644043
Train... ===> epoch = 95, loss = 3.773008346557617
Train... ===> epoch = 96, loss = 3.7673189640045166
Train... ===> epoch = 97, loss = 3.761645555496216
Train... ===> epoch = 98, loss = 3.755985736846924
Train... ===> epoch = 99, loss = 3.7503416538238525

绘制曲线：epoch与loss

plt.plot(epochs, loss_list)
plt.xlabel("epoch")
plt.ylabel("loss_val")
plt.show()

查看权重、偏置信息

print(f"w = {my_model.linear.weight}")
print(f"b = {my_model.linear.bias}")

w = Parameter containing:
tensor([[0.1506, 0.0253]], requires_grad=True)
b = Parameter containing:
tensor([-0.0544], requires_grad=True)

利用模型做预测

• 代码

X_test = torch.Tensor([[4.0, 12.0]])
y_pred = my_model(X_test)
print(f"y_pred = {y_pred.data}")

• 输出信息

y_pred = tensor([[0.7010]])

• 代码

import numpy as np

x = np.linspace(0, 10, 40)
x_t = torch.Tensor(x).view((20, 2))
y_t = my_model(x_t)
y = y_t.data.numpy()

for xe,ye in zip(x_t, y_t):
    print(xe.data, '\t', ye.data)

• 输出信息

tensor([0.0000, 0.2564]) 	 tensor([0.4880])
tensor([0.5128, 0.7692]) 	 tensor([0.5106])
tensor([1.0256, 1.2821]) 	 tensor([0.5331])
tensor([1.5385, 1.7949]) 	 tensor([0.5554])
tensor([2.0513, 2.3077]) 	 tensor([0.5776])
tensor([2.5641, 2.8205]) 	 tensor([0.5994])
tensor([3.0769, 3.3333]) 	 tensor([0.6209])
tensor([3.5897, 3.8462]) 	 tensor([0.6419])
tensor([4.1026, 4.3590]) 	 tensor([0.6623])
tensor([4.6154, 4.8718]) 	 tensor([0.6822])
tensor([5.1282, 5.3846]) 	 tensor([0.7014])
tensor([5.6410, 5.8974]) 	 tensor([0.7200])
tensor([6.1538, 6.4103]) 	 tensor([0.7378])
tensor([6.6667, 6.9231]) 	 tensor([0.7549])
tensor([7.1795, 7.4359]) 	 tensor([0.7712])
tensor([7.6923, 7.9487]) 	 tensor([0.7867])
tensor([8.2051, 8.4615]) 	 tensor([0.8015])
tensor([8.7179, 8.9744]) 	 tensor([0.8154])
tensor([9.2308, 9.4872]) 	 tensor([0.8286])
tensor([ 9.7436, 10.0000]) 	 tensor([0.8410])