【pytorch】手动实现多层感知机（MultiLayerPerceptron）（torch.max(input=X, other=torch.tensor(0.0))的分析+多层感知机的简单实现
文章目录
- 一、手写函数
- （1）代码
- （2）结果
- 二、一点思考
- （1）关于ReLu的思考
- （2）关于学习率lr的思考
- 1）原因思考
- 2）改正代码
- 三、利用pytorch模块简单实现多层感知机
- （1）代码
- （2）结果
- 参考
一、手写函数
（1）代码
import sys
import numpy as np
import torch
import torchvision
def sgd(params, lr, batch_size):
# 为了和原书保持一致，这里除以了batch_size，但是应该是不用除的，因为一般用PyTorch计算loss时就默认已经
# 沿batch维求了平均了。
for param in params:
param.data -= lr * param.grad / batch_size # 注意这里更改param时用的param.data
def evaluate_accuracy(data_iter, net, device=None):
if device is None and isinstance(net, torch.nn.Module):
# 如果没指定device就使用net的device
device = list(net.parameters())[0].device
acc_sum, n = 0.0, 0
with torch.no_grad():
for X, y in data_iter:
if isinstance(net, torch.nn.Module):
net.eval() # 评估模式, 这会关闭dropout
acc_sum += (net(X.to(device)).argmax(dim=1) == y.to(device)).float().sum().cpu().item()
net.train() # 改回训练模式
else: # 自定义的模型, 3.13节之后不会用到, 不考虑GPU
if ('is_training' in net.__code__.co_varnames): # 如果有is_training这个参数
# 将is_training设置成False
acc_sum += (net(X, is_training=False).argmax(dim=1) == y).float().sum().item()
else:
acc_sum += (net(X).argmax(dim=1) == y).float().sum().item()
n += y.shape[0]
return acc_sum / n
def train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size,
params=None, lr=None, optimizer=None):
for epoch in range(num_epochs):
train_l_sum, train_acc_sum, n = 0.0, 0.0, 0
for X, y in train_iter:
y_hat = net(X)
l = loss(y_hat, y).sum()
# 梯度清零
if optimizer is not None:
optimizer.zero_grad()
elif params is not None and params[0].grad is not None:
for param in params:
param.grad.data.zero_()
l.backward()
if optimizer is None:
sgd(params, lr, batch_size)
else:
optimizer.step() # “softmax回归的简洁实现”一节将用到
train_l_sum += l.item()
train_acc_sum += (y_hat.argmax(dim=1) == y).sum().item()
n += y.shape[0]
test_acc = evaluate_accuracy(test_iter, net)
print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f'
% (epoch + 1, train_l_sum / n, train_acc_sum / n, test_acc))
def load_data_fashion_mnist(batch_size, resize=None, root='./data'):
"""Download the fashion mnist dataset and then load into memory."""
trans = []
if resize:
trans.append(torchvision.transforms.Resize(size=resize))
trans.append(torchvision.transforms.ToTensor())
transform = torchvision.transforms.Compose(trans)
mnist_train = torchvision.datasets.FashionMNIST(root=root, train=True, download=True, transform=transform)
mnist_test = torchvision.datasets.FashionMNIST(root=root, train=False, download=True, transform=transform)
if sys.platform.startswith('win'):
num_workers = 0 # 0表示不用额外的进程来加速读取数据
else:
num_workers = 4
train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)
return train_iter, test_iter
def relu(X):
return torch.max(input=X, other=torch.tensor(0.0))
def net(X):
X = X.view((-1, num_inputs))
H = relu(torch.matmul(X, W1) + b1)
return torch.matmul(H, W2) + b2
if __name__ == '__main__':
batch_size = 256
train_iter, test_iter = load_data_fashion_mnist(batch_size)
num_inputs, num_outputs, num_hiddens = 784, 10, 256
W1 = torch.tensor(np.random.normal(0, 0.01, (num_inputs, num_hiddens)), dtype=torch.float)
b1 = torch.zeros(num_hiddens, dtype=torch.float)
W2 = torch.tensor(np.random.normal(0, 0.01, (num_hiddens, num_outputs)), dtype=torch.float)
b2 = torch.zeros(num_outputs, dtype=torch.float)
params = [W1, b1, W2, b2]
for param in params:
param.requires_grad_(requires_grad=True)
loss = torch.nn.CrossEntropyLoss()
num_epochs, lr = 5, 100.0
train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, params, lr)
（2）结果
epoch 1, loss 0.0031, train acc 0.710, test acc 0.798
epoch 2, loss 0.0019, train acc 0.823, test acc 0.797
epoch 3, loss 0.0017, train acc 0.845, test acc 0.846
epoch 4, loss 0.0015, train acc 0.854, test acc 0.840
epoch 5, loss 0.0015, train acc 0.863, test acc 0.849
二、一点思考
（1）关于ReLu的思考
- ReLu函数中的
torch.max(input=X, other=torch.tensor(0.0))这句函数中other参数的意义我不是很清楚，所以我写了一个测试代码来分析：
import numpy as np
import torch
def ReLu(x):
return torch.max(input=x, other=torch.tensor(0.0))
x = torch.tensor(np.random.normal(0, 0.1, (3, 3)), dtype=torch.float)
y = ReLu(x)
print(x, '\n', y)
- 结果
tensor([[-0.0069, 0.2676, 0.1593],
[ 0.2727, -0.0243, 0.0467],
[-0.1764, -0.1287, -0.0419]])
tensor([[0.0000, 0.2676, 0.1593],
[0.2727, 0.0000, 0.0467],
[0.0000, 0.0000, 0.0000]])
- 分析：可以发现这句代码完美的实现了ReLu函数应有的功能，其中other参数的意义应该是给定一个tensor，在输入x中，若比给定的tensor小，则将值替换为改tensor，否则不变的意思。
（2）关于学习率lr的思考
1）原因思考
- 阅读上面函数时可以发现112行的lr=100.0，引起了我的思考，为什么lr这么大？
def sgd(params, lr, batch_size):
# 为了和原书保持一致，这里除以了batch_size，但是应该是不用除的，因为一般用PyTorch计算loss时就默认已经
# 沿batch维求了平均了。
for param in params:
param.data -= lr * param.grad / batch_size # 注意这里更改param时用的param.data
- 原因在上面这段代码里面，由于本代码里面loss使用的是官方的loss函数：
loss = torch.nn.CrossEntropyLoss()，默认已经沿batch维求了平均了，在计算参数回传的时候，又除了一遍batch_size，就导致了回传参数小了百倍，所以如果用lr=0.1来训练的话，epoch准确率如下，很低。正确的应该不要再除以batch_size即可。
epoch 1, loss 0.0090, train acc 0.059, test acc 0.122
epoch 2, loss 0.0090, train acc 0.218, test acc 0.276
epoch 3, loss 0.0090, train acc 0.286, test acc 0.291
epoch 4, loss 0.0090, train acc 0.297, test acc 0.293
epoch 5, loss 0.0089, train acc 0.298, test acc 0.294
- sgd函数最开始定义在【pytorch】手写softmax回归单层网络模型范例 + 利用pytorch模块简单实现softmax回归中，在那篇博客里面，loss和参数回传的sgd函数要么都是手写的，要么都是官方的，都只除了一遍，所以就没有问题，但是在本文中，loss是官方，sgd是手写，所以就出现了问题。
2）改正代码
import sys
import numpy as np
import torch
import torchvision
def sgd(params, lr, batch_size):
# 为了和原书保持一致，这里除以了batch_size，但是应该是不用除的，因为一般用PyTorch计算loss时就默认已经
# 沿batch维求了平均了。
for param in params:
param.data -= lr * param.grad # 注意这里更改param时用的param.data
def evaluate_accuracy(data_iter, net, device=None):
if device is None and isinstance(net, torch.nn.Module):
# 如果没指定device就使用net的device
device = list(net.parameters())[0].device
acc_sum, n = 0.0, 0
with torch.no_grad():
for X, y in data_iter:
if isinstance(net, torch.nn.Module):
net.eval() # 评估模式, 这会关闭dropout
acc_sum += (net(X.to(device)).argmax(dim=1) == y.to(device)).float().sum().cpu().item()
net.train() # 改回训练模式
else: # 自定义的模型, 3.13节之后不会用到, 不考虑GPU
if ('is_training' in net.__code__.co_varnames): # 如果有is_training这个参数
# 将is_training设置成False
acc_sum += (net(X, is_training=False).argmax(dim=1) == y).float().sum().item()
else:
acc_sum += (net(X).argmax(dim=1) == y).float().sum().item()
n += y.shape[0]
return acc_sum / n
def train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size,
params=None, lr=None, optimizer=None):
for epoch in range(num_epochs):
train_l_sum, train_acc_sum, n = 0.0, 0.0, 0
for X, y in train_iter:
y_hat = net(X)
l = loss(y_hat, y).sum()
# 梯度清零
if optimizer is not None:
optimizer.zero_grad()
elif params is not None and params[0].grad is not None:
for param in params:
param.grad.data.zero_()
l.backward()
if optimizer is None:
sgd(params, lr, batch_size)
else:
optimizer.step() # “softmax回归的简洁实现”一节将用到
train_l_sum += l.item()
train_acc_sum += (y_hat.argmax(dim=1) == y).sum().item()
n += y.shape[0]
test_acc = evaluate_accuracy(test_iter, net)
print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f'
% (epoch + 1, train_l_sum / n, train_acc_sum / n, test_acc))
def load_data_fashion_mnist(batch_size, resize=None, root='./data'):
"""Download the fashion mnist dataset and then load into memory."""
trans = []
if resize:
trans.append(torchvision.transforms.Resize(size=resize))
trans.append(torchvision.transforms.ToTensor())
transform = torchvision.transforms.Compose(trans)
mnist_train = torchvision.datasets.FashionMNIST(root=root, train=True, download=True, transform=transform)
mnist_test = torchvision.datasets.FashionMNIST(root=root, train=False, download=True, transform=transform)
if sys.platform.startswith('win'):
num_workers = 0 # 0表示不用额外的进程来加速读取数据
else:
num_workers = 4
train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)
return train_iter, test_iter
def relu(X):
return torch.max(input=X, other=torch.tensor(0.0))
def net(X):
X = X.view((-1, num_inputs))
H = relu(torch.matmul(X, W1) + b1)
return torch.matmul(H, W2) + b2
if __name__ == '__main__':
batch_size = 256
train_iter, test_iter = load_data_fashion_mnist(batch_size)
num_inputs, num_outputs, num_hiddens = 784, 10, 256
W1 = torch.tensor(np.random.normal(0, 0.01, (num_inputs, num_hiddens)), dtype=torch.float)
b1 = torch.zeros(num_hiddens, dtype=torch.float)
W2 = torch.tensor(np.random.normal(0, 0.01, (num_hiddens, num_outputs)), dtype=torch.float)
b2 = torch.zeros(num_outputs, dtype=torch.float)
params = [W1, b1, W2, b2]
for param in params:
param.requires_grad_(requires_grad=True)
loss = torch.nn.CrossEntropyLoss()
num_epochs, lr = 5, 0.1
train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, params, lr)
- 结果
epoch 1, loss 0.0041, train acc 0.634, test acc 0.703
epoch 2, loss 0.0023, train acc 0.790, test acc 0.807
epoch 3, loss 0.0020, train acc 0.819, test acc 0.778
epoch 4, loss 0.0019, train acc 0.830, test acc 0.800
epoch 5, loss 0.0018, train acc 0.840, test acc 0.828
三、利用pytorch模块简单实现多层感知机
（1）代码
import sys
import torch
import torch.nn as nn
import torchvision
from torch.nn import init
class FlattenLayer(nn.Module):
def __init__(self):
super(FlattenLayer, self).__init__()
def forward(self, x): # x shape: (batch, *, *, ...)
return x.view(x.shape[0], -1)
def sgd(params, lr, batch_size):
# 为了和原书保持一致，这里除以了batch_size，但是应该是不用除的，因为一般用PyTorch计算loss时就默认已经
# 沿batch维求了平均了。
for param in params:
param.data -= lr * param.grad # 注意这里更改param时用的param.data
def evaluate_accuracy(data_iter, net, device=None):
if device is None and isinstance(net, torch.nn.Module):
# 如果没指定device就使用net的device
device = list(net.parameters())[0].device
acc_sum, n = 0.0, 0
with torch.no_grad():
for X, y in data_iter:
if isinstance(net, torch.nn.Module):
net.eval() # 评估模式, 这会关闭dropout
acc_sum += (net(X.to(device)).argmax(dim=1) == y.to(device)).float().sum().cpu().item()
net.train() # 改回训练模式
else: # 自定义的模型, 3.13节之后不会用到, 不考虑GPU
if ('is_training' in net.__code__.co_varnames): # 如果有is_training这个参数
# 将is_training设置成False
acc_sum += (net(X, is_training=False).argmax(dim=1) == y).float().sum().item()
else:
acc_sum += (net(X).argmax(dim=1) == y).float().sum().item()
n += y.shape[0]
return acc_sum / n
def train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size,
params=None, lr=None, optimizer=None):
for epoch in range(num_epochs):
train_l_sum, train_acc_sum, n = 0.0, 0.0, 0
for X, y in train_iter:
y_hat = net(X)
l = loss(y_hat, y).sum()
# 梯度清零
if optimizer is not None:
optimizer.zero_grad()
elif params is not None and params[0].grad is not None:
for param in params:
param.grad.data.zero_()
l.backward()
if optimizer is None:
sgd(params, lr, batch_size)
else:
optimizer.step() # “softmax回归的简洁实现”一节将用到
train_l_sum += l.item()
train_acc_sum += (y_hat.argmax(dim=1) == y).sum().item()
n += y.shape[0]
test_acc = evaluate_accuracy(test_iter, net)
print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f'
% (epoch + 1, train_l_sum / n, train_acc_sum / n, test_acc))
def load_data_fashion_mnist(batch_size, resize=None, root='./data'):
"""Download the fashion mnist dataset and then load into memory."""
trans = []
if resize:
trans.append(torchvision.transforms.Resize(size=resize))
trans.append(torchvision.transforms.ToTensor())
transform = torchvision.transforms.Compose(trans)
mnist_train = torchvision.datasets.FashionMNIST(root=root, train=True, download=True, transform=transform)
mnist_test = torchvision.datasets.FashionMNIST(root=root, train=False, download=True, transform=transform)
if sys.platform.startswith('win'):
num_workers = 0 # 0表示不用额外的进程来加速读取数据
else:
num_workers = 4
train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)
return train_iter, test_iter
if __name__ == '__main__':
num_inputs, num_outputs, num_hiddens = 784, 10, 256
net = nn.Sequential(
FlattenLayer(),
nn.Linear(num_inputs, num_hiddens),
nn.ReLU(),
nn.Linear(num_hiddens, num_outputs),
)
for params in net.parameters():
init.normal_(params, mean=0, std=0.01)
batch_size = 256
train_iter, test_iter = load_data_fashion_mnist(batch_size)
loss = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr=0.1)
num_epochs = 5
train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, None, None, optimizer)
（2）结果
epoch 1, loss 0.0041, train acc 0.639, test acc 0.701
epoch 2, loss 0.0024, train acc 0.789, test acc 0.770
epoch 3, loss 0.0020, train acc 0.819, test acc 0.799
epoch 4, loss 0.0019, train acc 0.831, test acc 0.805
epoch 5, loss 0.0018, train acc 0.842, test acc 0.823
参考
动手学深度学习（pytorch版）