当前位置：首页 > news >正文

基于Fashion-MNIST的softmax回归-直接运行

news 来源：原创 2025/5/18 7:06:56

引用 Fashion-MNIST数据集，进行分类问题训练，代码如下，可直接运行

import torch
import torchvision
from torchvision import transforms
from torch.utils import data
import timeclass Timer:"""记录多次运行时间"""def __init__(self):"""Defined in :numref:`subsec_linear_model`"""self.times = []self.start()def start(self):"""启动计时器"""self.tik = time.time()def stop(self):"""停止计时器并将时间记录在列表中"""self.times.append(time.time() - self.tik)return self.times[-1]def avg(self):"""返回平均时间"""return sum(self.times) / len(self.times)def sum(self):"""返回时间总和"""return sum(self.times)def cumsum(self):"""返回累计时间"""return np.array(self.times).cumsum().tolist()size = lambda x, *args, **kwargs: x.numel(*args, **kwargs)
reduce_sum = lambda x, *args, **kwargs: x.sum(*args, **kwargs)
argmax = lambda x, *args, **kwargs: x.argmax(*args, **kwargs)
astype = lambda x, *args, **kwargs: x.type(*args, **kwargs)def accuracy(y_hat, y):"""计算预测正确的数量"""if len(y_hat.shape) > 1 and y_hat.shape[1] > 1:y_hat = argmax(y_hat, axis=1)cmp = astype(y_hat, y.dtype) == yreturn float(reduce_sum(astype(cmp, y.dtype)))def evaluate_accuracy(net, data_iter):"""计算在指定数据集上模型的精度"""metric = Accumulator(2)  # 正确预测数、预测总数with torch.no_grad():for X, y in data_iter:metric.add(accuracy(net(X), y), size(y))return metric[0] / metric[1]class Accumulator:"""在n个变量上累加"""def __init__(self, n):self.data = [0.0] * ndef add(self, *args):self.data = [a + float(b) for a, b in zip(self.data, args)]def reset(self):self.data = [0.0] * len(self.data)def __getitem__(self, idx):return self.data[idx]def get_dataloader_workers():return 4def load_data_fashion_mnist(batch_size, resize=None):"""下载Fashion-MNIST数据集，然后将其加载到内存中"""trans = [transforms.ToTensor()]if resize:trans.insert(0, transforms.Resize(resize))trans = transforms.Compose(trans)mnist_train = torchvision.datasets.FashionMNIST(root="../data", train=True, transform=trans, download=True)mnist_test = torchvision.datasets.FashionMNIST(root="../data", train=False, transform=trans, download=True)return (data.DataLoader(mnist_train, batch_size, shuffle=True,num_workers=get_dataloader_workers()),data.DataLoader(mnist_test, batch_size, shuffle=False,num_workers=get_dataloader_workers()))def softmax(X):X_exp = torch.exp(X)partition = X_exp.sum(1, keepdim=True)return X_exp / partitiondef net(X):return softmax(torch.matmul(X.reshape((-1, W.shape[0])), W) + b)def cross_entropy(y_hat, y):#get the prediction probability b y，从表面上看，损失与类别数量无关probability = y_hat[range(len(y_hat)), y]return -torch.log(probability)def train_epoch_ch3(net, train_iter, loss, updater):# 训练损失总和、训练准确度总和、样本数metric = Accumulator(3)for X, y in train_iter:# 计算梯度并更新参数y_hat = net(X)l = loss(y_hat, y)# 使用定制的优化器和损失函数l.sum().backward()sgd([W,b], lr, X.shape[0])#batch_sizemetric.add(float(l.sum()), accuracy(y_hat, y), y.numel())# 返回训练损失和训练精度return metric[0] / metric[2], metric[1] / metric[2]def train_ch3(net, train_iter, test_iter, loss, num_epochs, updater):for epoch in range(num_epochs):train_metrics = train_epoch_ch3(net, train_iter, loss, updater)test_acc = evaluate_accuracy(net, test_iter)print(f'epoch {epoch + 1}, loss={train_metrics[0]:.5f}, train_acc={train_metrics[1]:.5f}, test_acc={test_acc:.5f}')return train_metricsdef sgd(params, lr, batch_size):with torch.no_grad():for param in params:param -= lr * param.grad / batch_sizeparam.grad.zero_()lr = 0.1
num_epochs = 10
batch_size = 256num_inputs = 784
num_outputs = 10W = torch.normal(0, 0.01, size=(num_inputs, num_outputs), requires_grad=True)
b = torch.zeros(num_outputs, requires_grad=True)train_iter, test_iter = load_data_fashion_mnist(batch_size)
timer = Timer()
train_loss, train_acc = train_ch3(net, train_iter, test_iter, cross_entropy, num_epochs, sgd)
print(f'train takes {timer.stop():.2f} sec, loss={train_loss:.5f}, train_acc={train_acc:.5f}')

运行结果：

epoch 1, loss=0.78627, train_acc=0.74832, test_acc=0.79040
epoch 2, loss=0.57115, train_acc=0.81300, test_acc=0.81000
epoch 3, loss=0.52513, train_acc=0.82650, test_acc=0.81280
epoch 4, loss=0.50200, train_acc=0.83142, test_acc=0.82520
epoch 5, loss=0.48449, train_acc=0.83667, test_acc=0.82270
epoch 6, loss=0.47343, train_acc=0.83960, test_acc=0.82230
epoch 7, loss=0.46612, train_acc=0.84238, test_acc=0.82810
epoch 8, loss=0.45860, train_acc=0.84573, test_acc=0.82260
epoch 9, loss=0.45168, train_acc=0.84740, test_acc=0.83220
epoch 10, loss=0.44680, train_acc=0.84780, test_acc=0.83150
train takes 84.08 sec, loss=0.44680, train_acc=0.84780

训练10次，可以看到，总耗时84秒（CPU），后面改为使用GPU时间会短一些，后面再结合另外一篇文章一块整合；随着训练的进行，损失函数逐渐减小，准确率逐渐增大。