Starter Project

%matplotlib inline from torch import nn from IPython import display from torch.utils.data import TensorDataset, DataLoader import torchvision import torch import torchvision.transforms as transforms import gzip import numpy as np import time import os import matplotlib.pyplot as plt display.set_matplotlib_formats("svg")

mnist_train = torchvision.datasets.FashionMNIST(root='../dataset/fashion_mnist', train=True, download=True, transform=transforms.ToTensor()) mnist_test = torchvision.datasets.FashionMNIST(root='../dataset/fashion_mnist', train=False, download=True, transform=transforms.ToTensor())

feature, label = mnist_train[0] print(feature.shape, label) # Channel x Height x Width

def get_fashion_mnist_labels(labels): text_labels = ['t-shirt', 'trouser', 'pullover', 'dress', 'coat', 'sandal', 'shirt', 'sneaker', 'bag', 'ankle boot'] return [text_labels[int(i)] for i in labels]

def show_fashion_mnist(images, labels): _, figs = plt.subplots(1, len(images), figsize=(12, 12)) for f, img, lbl in zip(figs, images, labels): f.imshow(img.view((28, 28)).numpy()) f.set_title(lbl) f.axes.get_xaxis().set_visible(False) f.axes.get_yaxis().set_visible(False) plt.show()

X, y = [], [] for i in range(10): X.append(mnist_train[i][0]) y.append(mnist_train[i][1]) show_fashion_mnist(X, get_fashion_mnist_labels(y))

batch_size = 256 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)

def accuracy(net, data_iter, is_one_hot=False): ''' net: 神经网络 data_iter: 数据迭代器 ''' sum_acc = 0.0 sum_crossEnt = 0.0 n = 0 for X, y in data_iter: hat_y = torch.softmax(net(X), dim=1) if is_one_hot: acc = torch.sum(torch.max(y, axis=1)[1] == torch.max(hat_y, axis=1)[1]).item() crossEnt = -torch.dot(y.float(), torch.log2(hat_y)).item() else: acc = torch.sum((y == torch.max(hat_y, axis=1)[1]).byte()).numpy().item() crossEnt = -torch.dot(y.float(), torch.log2(hat_y[torch.arange(hat_y.shape[0]), y.long()].reshape(-1))).item() sum_acc += acc sum_crossEnt += crossEnt n += X.shape[0] return sum_crossEnt/n, sum_acc/n

class FlattenLayer(torch.nn.Module): def __init__(self): super(FlattenLayer, self).__init__() def forward(self, x): # x shape: (batch, *, *, ...) return x.view(x.shape[0], -1)

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(): torch.nn.init.normal_(params, mean=0, std=0.001) loss = torch.nn.CrossEntropyLoss() optimizer = torch.optim.SGD(net.parameters(), lr=0.1)

t1 = time.perf_counter() num_epochs = 40 for epoch in range(num_epochs): for t_x, t_y in train_iter: l = loss(net(t_x), t_y) # 计算当前批量的交叉熵损失 optimizer.zero_grad() # 参数梯度清零 l.backward() # 反向传播，计算梯度 optimizer.step() # 更新参数 with torch.no_grad(): # 不计算梯度，加速损失函数的运算 train_accuracy = accuracy(net, train_iter) test_accuracy = accuracy(net, test_iter) print(f'Epoch {epoch+1}, train_loss: {train_accuracy[0]:.4f} train_accuracy: {train_accuracy[1]:.4f}, test_loss: {test_accuracy[0]:.4f}, test_accuracy: {test_accuracy[1]:.4f}') print(f"耗时:{time.perf_counter()-t1:.2f}s")