PyTorch深度学习实践（五）课后作业

转载

mb5fed440247fb5 2021-08-03 23:22:00

线性模型基础，引自视频：刘二大人——《PyTorch深度学习实践》完结合集（p5）

要求

绘制各种优化算法的曲线进行比较
以训练次数为x，loss为y进行绘制

代码如下：

点击查看代码

import torch
import matplotlib.pyplot as plt

# 原始数据
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)  # 第一个1为输入维度，第二个1为输出维度

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


# 使用不同的优化方法进行训练，str为训练方法
def train(model, optimizer, str):
    loss_list = []
    criterion = torch.nn.MSELoss(size_average=False)
    for epoch in range(100):
        y_pred = model(x_data)
        loss = criterion(y_pred, y_data)
        # print(epoch, loss.item())
        loss_list.append(loss.item())
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

    # 绘制曲线
    epoch = range(len(loss_list))
    plt.plot(epoch, loss_list, label=str)


# SGD曲线
model = LinearModel()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
train(model, optimizer, 'SGD')

# Adam曲线
model = LinearModel()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
train(model, optimizer, 'Adam')

# Adagrad曲线
model = LinearModel()
optimizer = torch.optim.Adagrad(model.parameters(), lr=0.01)
train(model, optimizer, 'Adagrad')

# Adamax曲线
model = LinearModel()
optimizer = torch.optim.Adamax(model.parameters(), lr=0.01)
train(model, optimizer, 'Adamax')

# ASGD曲线
model = LinearModel()
optimizer = torch.optim.ASGD(model.parameters(), lr=0.01)
train(model, optimizer, 'ASGD')

# LBFGS曲线
# model = LinearModel()
# optimizer = torch.optim.LBFGS(model.parameters(), lr=0.01)
# train(model, optimizer, 'LBFGS')

# RMSprop
model = LinearModel()
optimizer = torch.optim.RMSprop(model.parameters(), lr=0.01)
train(model, optimizer, 'RMSprop')

# Rprop曲线
model = LinearModel()
optimizer = torch.optim.Rprop(model.parameters(), lr=0.01)
train(model, optimizer, 'Rprop')

plt.legend()
plt.show()
# print('w=', model.linear.weight.item())
# print('b=', model.linear.bias.item())
#
x_test = torch.Tensor([4.0])
y_test = model(x_test)
print('y_test=', y_test.data)