pytorch-05. Implementing linear regression with pytorch

import torch

x_data = torch.Tensor([[1.0],[2.0],[3.0]])
y_data = torch.Tensor([[2.0],[4.0],[6.0]])

class LinearModel(torch.nn.Module):
    def __init__(self):
        super(LinearModel,self).__init__()
        self.linear = torch.nn.Linear(1,1) #输入维度为1 输出维度为1

    def forward(self, x):
        y_pred  = self.linear(x)
        return y_pred

model = LinearModel()

criterion = torch.nn.MSELoss(size_average=True)   #损失函数
optimizer = torch.optim.SGD(model.parameters(),lr=0.01)   #优化器
'''
求解过程：
求y^
loss
梯度清零
backward
更新
'''
for epoch in range(100):
    y_pred = model(x_data)
    loss = criterion(y_pred,y_data)
    print('epoch=',epoch," loss=",loss)

    optimizer.zero_grad()  #梯度归0
    loss.backward()    #反向传播
    optimizer.step()   #更新

# Output weight and bias
print('w=',model.linear.weight.item())
print('b=',model.linear.bias.item())

# Test Model
x_test = torch.Tensor([[4.0]])
y_test = model(x_test)
#两种输出格式
print('y_pred=',y_test.data)
print('y_pred=',y_test.data.item())

结果：

epoch= 0  loss= tensor(34.0210, grad_fn=<MseLossBackward0>)
epoch= 1  loss= tensor(26.9701, grad_fn=<MseLossBackward0>)
epoch= 2  loss= tensor(21.3965, grad_fn=<MseLossBackward0>)
epoch= 3  loss= tensor(16.9904, grad_fn=<MseLossBackward0>)
epoch= 4  loss= tensor(13.5073, grad_fn=<MseLossBackward0>)
epoch= 5  loss= tensor(10.7537, grad_fn=<MseLossBackward0>)
epoch= 6  loss= tensor(8.5767, grad_fn=<MseLossBackward0>)
epoch= 7  loss= tensor(6.8556, grad_fn=<MseLossBackward0>)
epoch= 8  loss= tensor(5.4948, grad_fn=<MseLossBackward0>)
epoch= 9  loss= tensor(4.4188, grad_fn=<MseLossBackward0>)
epoch= 10  loss= tensor(3.5679, grad_fn=<MseLossBackward0>)
epoch= 11  loss= tensor(2.8949, grad_fn=<MseLossBackward0>)
epoch= 12  loss= tensor(2.3626, grad_fn=<MseLossBackward0>)
epoch= 13  loss= tensor(1.9415, grad_fn=<MseLossBackward0>)
epoch= 14  loss= tensor(1.6083, grad_fn=<MseLossBackward0>)
epoch= 15  loss= tensor(1.3446, grad_fn=<MseLossBackward0>)
epoch= 16  loss= tensor(1.1357, grad_fn=<MseLossBackward0>)
epoch= 17  loss= tensor(0.9703, grad_fn=<MseLossBackward0>)
epoch= 18  loss= tensor(0.8392, grad_fn=<MseLossBackward0>)
epoch= 19  loss= tensor(0.7352, grad_fn=<MseLossBackward0>)
epoch= 20  loss= tensor(0.6527, grad_fn=<MseLossBackward0>)
epoch= 21  loss= tensor(0.5871, grad_fn=<MseLossBackward0>)
epoch= 22  loss= tensor(0.5350, grad_fn=<MseLossBackward0>)
epoch= 23  loss= tensor(0.4934, grad_fn=<MseLossBackward0>)
epoch= 24  loss= tensor(0.4602, grad_fn=<MseLossBackward0>)
epoch= 25  loss= tensor(0.4336, grad_fn=<MseLossBackward0>)
epoch= 26  loss= tensor(0.4122, grad_fn=<MseLossBackward0>)
epoch= 27  loss= tensor(0.3950, grad_fn=<MseLossBackward0>)
epoch= 28  loss= tensor(0.3811, grad_fn=<MseLossBackward0>)
epoch= 29  loss= tensor(0.3697, grad_fn=<MseLossBackward0>)
epoch= 30  loss= tensor(0.3604, grad_fn=<MseLossBackward0>)
epoch= 31  loss= tensor(0.3527, grad_fn=<MseLossBackward0>)
epoch= 32  loss= tensor(0.3464, grad_fn=<MseLossBackward0>)
epoch= 33  loss= tensor(0.3410, grad_fn=<MseLossBackward0>)
epoch= 34  loss= tensor(0.3364, grad_fn=<MseLossBackward0>)
epoch= 35  loss= tensor(0.3325, grad_fn=<MseLossBackward0>)
epoch= 36  loss= tensor(0.3290, grad_fn=<MseLossBackward0>)
epoch= 37  loss= tensor(0.3260, grad_fn=<MseLossBackward0>)
epoch= 38  loss= tensor(0.3233, grad_fn=<MseLossBackward0>)
epoch= 39  loss= tensor(0.3208, grad_fn=<MseLossBackward0>)
epoch= 40  loss= tensor(0.3186, grad_fn=<MseLossBackward0>)
epoch= 41  loss= tensor(0.3165, grad_fn=<MseLossBackward0>)
epoch= 42  loss= tensor(0.3145, grad_fn=<MseLossBackward0>)
epoch= 43  loss= tensor(0.3126, grad_fn=<MseLossBackward0>)
epoch= 44  loss= tensor(0.3109, grad_fn=<MseLossBackward0>)
epoch= 45  loss= tensor(0.3092, grad_fn=<MseLossBackward0>)
epoch= 46  loss= tensor(0.3075, grad_fn=<MseLossBackward0>)
epoch= 47  loss= tensor(0.3059, grad_fn=<MseLossBackward0>)
epoch= 48  loss= tensor(0.3043, grad_fn=<MseLossBackward0>)
epoch= 49  loss= tensor(0.3028, grad_fn=<MseLossBackward0>)
epoch= 50  loss= tensor(0.3012, grad_fn=<MseLossBackward0>)
epoch= 51  loss= tensor(0.2997, grad_fn=<MseLossBackward0>)
epoch= 52  loss= tensor(0.2983, grad_fn=<MseLossBackward0>)
epoch= 53  loss= tensor(0.2968, grad_fn=<MseLossBackward0>)
epoch= 54  loss= tensor(0.2953, grad_fn=<MseLossBackward0>)
epoch= 55  loss= tensor(0.2939, grad_fn=<MseLossBackward0>)
epoch= 56  loss= tensor(0.2925, grad_fn=<MseLossBackward0>)
epoch= 57  loss= tensor(0.2910, grad_fn=<MseLossBackward0>)
epoch= 58  loss= tensor(0.2896, grad_fn=<MseLossBackward0>)
epoch= 59  loss= tensor(0.2882, grad_fn=<MseLossBackward0>)
epoch= 60  loss= tensor(0.2869, grad_fn=<MseLossBackward0>)
epoch= 61  loss= tensor(0.2855, grad_fn=<MseLossBackward0>)
epoch= 62  loss= tensor(0.2841, grad_fn=<MseLossBackward0>)
epoch= 63  loss= tensor(0.2827, grad_fn=<MseLossBackward0>)
epoch= 64  loss= tensor(0.2814, grad_fn=<MseLossBackward0>)
epoch= 65  loss= tensor(0.2800, grad_fn=<MseLossBackward0>)
epoch= 66  loss= tensor(0.2787, grad_fn=<MseLossBackward0>)
epoch= 67  loss= tensor(0.2773, grad_fn=<MseLossBackward0>)
epoch= 68  loss= tensor(0.2760, grad_fn=<MseLossBackward0>)
epoch= 69  loss= tensor(0.2747, grad_fn=<MseLossBackward0>)
epoch= 70  loss= tensor(0.2733, grad_fn=<MseLossBackward0>)
epoch= 71  loss= tensor(0.2720, grad_fn=<MseLossBackward0>)
epoch= 72  loss= tensor(0.2707, grad_fn=<MseLossBackward0>)
epoch= 73  loss= tensor(0.2694, grad_fn=<MseLossBackward0>)
epoch= 74  loss= tensor(0.2681, grad_fn=<MseLossBackward0>)
epoch= 75  loss= tensor(0.2668, grad_fn=<MseLossBackward0>)
epoch= 76  loss= tensor(0.2656, grad_fn=<MseLossBackward0>)
epoch= 77  loss= tensor(0.2643, grad_fn=<MseLossBackward0>)
epoch= 78  loss= tensor(0.2630, grad_fn=<MseLossBackward0>)
epoch= 79  loss= tensor(0.2618, grad_fn=<MseLossBackward0>)
epoch= 80  loss= tensor(0.2605, grad_fn=<MseLossBackward0>)
epoch= 81  loss= tensor(0.2592, grad_fn=<MseLossBackward0>)
epoch= 82  loss= tensor(0.2580, grad_fn=<MseLossBackward0>)
epoch= 83  loss= tensor(0.2568, grad_fn=<MseLossBackward0>)
epoch= 84  loss= tensor(0.2555, grad_fn=<MseLossBackward0>)
epoch= 85  loss= tensor(0.2543, grad_fn=<MseLossBackward0>)
epoch= 86  loss= tensor(0.2531, grad_fn=<MseLossBackward0>)
epoch= 87  loss= tensor(0.2519, grad_fn=<MseLossBackward0>)
epoch= 88  loss= tensor(0.2507, grad_fn=<MseLossBackward0>)
epoch= 89  loss= tensor(0.2495, grad_fn=<MseLossBackward0>)
epoch= 90  loss= tensor(0.2483, grad_fn=<MseLossBackward0>)
epoch= 91  loss= tensor(0.2471, grad_fn=<MseLossBackward0>)
epoch= 92  loss= tensor(0.2459, grad_fn=<MseLossBackward0>)
epoch= 93  loss= tensor(0.2447, grad_fn=<MseLossBackward0>)
epoch= 94  loss= tensor(0.2435, grad_fn=<MseLossBackward0>)
epoch= 95  loss= tensor(0.2424, grad_fn=<MseLossBackward0>)
epoch= 96  loss= tensor(0.2412, grad_fn=<MseLossBackward0>)
epoch= 97  loss= tensor(0.2400, grad_fn=<MseLossBackward0>)
epoch= 98  loss= tensor(0.2389, grad_fn=<MseLossBackward0>)
epoch= 99  loss= tensor(0.2377, grad_fn=<MseLossBackward0>)
w= 1.4350532293319702
b= 1.2842093706130981
y_pred= tensor([[7.0244]])
y_pred= 7.0244221687316895

Process finished with exit code 0