python人工智能 DL MXNet深度学习: 用 MXNet 实现一个 DL 深度学习 任务，代码方案分享1(图文详解)

import mxnet as mx
from mxnet import gluon, nd
from mxnet.gluon import nn, data, loss

# 设置随机种子
mx.random.seed(42)

# 数据加载和预处理
train_data = data.vision.MNIST(train=True)
test_data = data.vision.MNIST(train=False)

transformer = data.vision.transforms.ToTensor()
train_data = train_data.transform_first(transformer)
test_data = test_data.transform_first(transformer)

batch_size = 64
train_loader = data.DataLoader(train_data, batch_size=batch_size, shuffle=True)
test_loader = data.DataLoader(test_data, batch_size=batch_size, shuffle=False)

# 定义模型
class CNN(nn.Block):
    def __init__(self, num_classes):
        super(CNN, self).__init__()
        self.features = nn.Sequential()
        with self.features.name_scope():
            self.features.add(
                nn.Conv2D(channels=32, kernel_size=(3, 3), activation='relu'),
                nn.MaxPool2D(pool_size=(2, 2)),
                nn.Conv2D(channels=64, kernel_size=(3, 3), activation='relu'),
                nn.MaxPool2D(pool_size=(2, 2)),
                nn.Flatten()
            )
        self.output = nn.Dense(num_classes)

    def forward(self, x):
        x = self.features(x)
        x = self.output(x)
        return x

# 初始化模型并定义损失函数和优化器
num_classes = 10
model = CNN(num_classes)
model.initialize(mx.init.Xavier())
criterion = loss.SoftmaxCrossEntropyLoss()
optimizer = mx.optimizer.Adam(learning_rate=0.001)

# 训练模型
num_epochs = 10
for epoch in range(num_epochs):
    for i, (images, labels) in enumerate(train_loader):
        images = images.as_in_context(mx.gpu())
        labels = labels.as_in_context(mx.gpu())

        with mx.autograd.record():
            outputs = model(images)
            loss = criterion(outputs, labels)

        loss.backward()
        optimizer.step(batch_size=batch_size)

        if (i + 1) % 100 == 0:
            print(f"Epoch [{epoch + 1}/{num_epochs}], Step [{i + 1}/{len(train_loader)}], Loss: {nd.mean(loss).asscalar()}")

# 在测试集上评估模型
model.collect_params().reset_ctx(mx.gpu())

total_correct = 0
total_samples = 0
for images, labels in test_loader:
    images = images.as_in_context(mx.gpu())
    labels = labels.as_in_context(mx.gpu())

    outputs = model(images)
    predicted_labels = nd.argmax(outputs, axis=1)
    total_samples += labels.shape[0]
    total_correct += (predicted_labels == labels).sum().asscalar()

accuracy = total_correct / total_samples
print(f"Test Accuracy: {accuracy}")