使用d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, trainer,)生成train loss图像

你可以按照以下步骤生成train loss图像：

1. 在train_ch3()函数的调用语句中添加参数eval_train=True，以便在训练过程中计算训练集上的损失值。例如：

   d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, trainer, eval_train=True)

2. 在train_ch3()函数的定义中，添加一个if语句，判断是否需要计算训练集上的损失值。例如：

   def train_ch3(net, train_iter, test_iter, loss, num_epochs, trainer, eval_train=False):
       for epoch in range(num_epochs):
           train_l_sum, train_acc_sum, n, m = 0.0, 0.0, 0, 0
           for X, y in train_iter:
               with autograd.record():
                   y_hat = net(X)
                   l = loss(y_hat, y).sum()
               l.backward()
               trainer.step(batch_size)
               train_l_sum += l.asscalar()
               train_acc_sum += (y_hat.argmax(axis=1) == y.astype('float32')).sum().asscalar()
               n += y.size
               m += y_hat.size
           test_acc = evaluate_accuracy(test_iter, net)
           if eval_train:
               train_acc = evaluate_accuracy(train_iter, net)
               print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f'
                     % (epoch + 1, train_l_sum / n, train_acc / m, test_acc))
           else:
               print('epoch %d, loss %.4f, test acc %.3f'
                     % (epoch + 1, train_l_sum / n, test_acc))

3. 在train_ch3()函数的调用语句之后，添加以下代码，生成train loss图像：

   train_loss = [i['train_loss'] for i in trainer.train_metrics]
   d2l.plot(range(1, num_epochs + 1), train_loss, xlabel='epoch', ylabel='loss')

这段代码会从trainer对象中提取出训练集的损失值，然后使用d2l.plot()函数生成train loss图像。

完整的代码如下：

import d2lzh as d2l
from mxnet import gluon, init, autograd
from mxnet.gluon import loss as gloss, nn

# 定义模型
net = nn.Sequential()
net.add(nn.Dense(256, activation='relu'),
        nn.Dense(10))
net.initialize(init.Normal(sigma=0.01))

# 定义损失函数
loss = gloss.SoftmaxCrossEntropyLoss()

# 定义优化算法
trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': 0.5})

# 训练模型
num_epochs = 10
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, trainer, eval_train=True)

# 生成train loss图像
train_loss = [i['train_loss'] for i in trainer.train_metrics]
d2l.plot(range(1, num_epochs + 1), train_loss, xlabel='epoch', ylabel='loss')

运行以上代码，就可以生成train loss图像。