学习参考:
10 多层感知机 + 代码实现 - 动手学深度学习v2_哔哩哔哩_bilibili
前序知识点:典型激活函数
A:Sigmoid激活函数
将输入投影至(0,1),是阶跃函数的soft版本
B:Tanh激活函数
将输入投影到(-1, 1)
C:ReLU激活函数
rectified linear unit
1. 多层感知机的从零开始实现
import torch
from torch import nn
from d2l import torch as d2l
d2l.load_data_fashion_mnist(batch_size, resize = None)
# 读取Fashion MNIST数据集,将其加载到内存中
# 通过转换列表,对数据进行格式转换,包括ToTensor和resize
# 调用内置数据加载器data.DataLoader()
d2l.get_dataloader_worker():
# 确定读取数据的进程数
d2l.accuracy(y_hat, y):
# 计算预测正确的样本数
d2l.evaluate_accuracy(net, data_iter):
# 计算指定模型在指定数据集上的精度
# 预测正确的样本数/当前批次总样本数
d2l.train_epoch_ch3(net, train_iter, loss, updater):
# 一个epoch内,更新模型参数
# 输出当前一个epoch训练完成后的精度和损失
d2l.train_ch3(net, train_iter, teat_iter, loss, num_epochs, updater)
# 完成所有epoch的训练,更新权重
# 可视化每轮epoch的train_loss,train_acc和test_acc
1.1 生成训练数据迭代生成器和测试/验证数据迭代生成器
batch_size = 256
train_iter, test_iter = load_data_fashion_mnist(batch)
1.2 定义多层感知机的结构和模型参数初始化
num_inputs, num_outputs, num_hiddens = 28 * 28, 10, 256
# num_inputs和num_outputs分别是输入特征维度,输出类别个数
# num_hiddens是隐藏层的大小(可以理解为隐藏层的特征维度)
W1 = nn.Parameter(
torch.randn(num_inputs, num_hiddens, requires_grad = True) * 0.01)
b1 = nn.Parameter(torch.zeros(num_hiddens), requires_grad = True)
W2 = nn.Parameter(
torch.randn(num_hiddens, num_outputs, requires_grad = True) * 0.01)
b2 = nn.Parameter(torch.zeros(num_outputs), requires_grad = True)
# 参数列表,用于更新梯度
params = [W1, b1, W2, b2]
1.3 实现ReLU函数
def relu(X):
a = torch.zeros_like(X)
# 生成一个和X大小相同的全0张量
return torch.max(X, a)
1.4 搭建模型
def net(X):
X = X.reshape(-1, num_inuts)
H = relu(X @ W1 + b1)
return (H @ W2 + b2)
1.5 多层感知机的训练过程
num_epochs, lr = 10, 0.1
loss = nn.CrossEntropy()
updater = torch.optim.SGD(params, lr = lr)
train_ch3(net, train_iter, test_iter, loss, num_epochs, updater)
1.6 在测试数据集应用模型
def predict_ch3(net, test_iter, n = 6):
for X, y in test_iter:
break
trues = d2l.get_fashion_mnist_labels(y)
preds = d2l.get_fashion_mnist_labels(net(X).argmax(axis = 1))
titles = [true + '\n' + pred for true, presd in zip(trues, pres)]
d2l.show_image(X[0:n].reshape((n, 28, 28)), 1, n, titles = titles[0: n])
predict_ch3(net, test_iter)
2. 多层感知机的简洁实现
import torch
from torch import nn
from d2l import torch as d2l
2.1 定义模型与初始化参数
net = nn.Sequential(
nn.Flatten(), nn.Linear(28*28, 256), nn.ReLU(), nn.Linear(256, 10))
def init_weights(m):
if type(m) == nn.Linear():
nn.init.normal_(m.weight, std = 0.01)
net.apply(init_weight)
2.2 训练过程
batch_size, lr, num_epoches = 256, 0.1, 10
loss = nn.CrossEntropy()
trainer = torch.optim.SGD(net.parameters(), lr = lr)
train_iter, test_iter = load_data_fashion_mnsit(batch_size)
train_ch3(net, train_iter, test_iter, loss, num_epoches, trainer)