sync lib

d2l/tensorflow.py

@@ -1352,10 +1352,8 @@ def train_concise_ch11(trainer_fn, hyperparams, data_iter, num_epochs=2):
     net.add(tf.keras.layers.Dense(1,
             kernel_initializer=tf.random_normal_initializer(stddev=0.01)))
     optimizer = trainer_fn(**hyperparams)
+    # 注意：MeanSquaredError计算平方误差时不带系数1/2
     loss = tf.keras.losses.MeanSquaredError()
-    # 注意:L2Loss=1/2*MSELoss。
-    # TensorFlow的MSE损失与MXNet的L2损失大概相差2倍。
-    # 因此，我们将TensorFlow中的损失减半
     animator = d2l.Animator(xlabel='epoch', ylabel='loss',
                             xlim=[0, num_epochs], ylim=[0.22, 0.35])
     n, timer = 0, d2l.Timer()
@@ -1363,7 +1361,7 @@ def train_concise_ch11(trainer_fn, hyperparams, data_iter, num_epochs=2):
         for X, y in data_iter:
             with tf.GradientTape() as g:
                 out = net(X)
-                l = loss(y, out)/2
+                l = loss(y, out)
                 params = net.trainable_variables
                 grads = g.gradient(l, params)
             optimizer.apply_gradients(zip(grads, params))
@@ -1372,7 +1370,7 @@ def train_concise_ch11(trainer_fn, hyperparams, data_iter, num_epochs=2):
                 timer.stop()
                 p = n/X.shape[0]
                 q = p/tf.data.experimental.cardinality(data_iter).numpy()
-                r = (d2l.evaluate_loss(net, data_iter, loss)/2,)
+                r = (d2l.evaluate_loss(net, data_iter, loss),)
                 animator.add(q, r)
                 timer.start()
     print(f'loss: {animator.Y[0][-1]:.3f}, {timer.avg():.3f} sec/epoch')

d2l/torch.py

@@ -259,15 +259,13 @@ def train_epoch_ch3(net, train_iter, loss, updater):
         if isinstance(updater, torch.optim.Optimizer):
             # 使用PyTorch内置的优化器和损失函数
             updater.zero_grad()
-            l.backward()
+            l.sum().backward()
             updater.step()
-            metric.add(float(l) * len(y), accuracy(y_hat, y),
-                       y.size().numel())
         else:
             # 使用定制的优化器和损失函数
             l.sum().backward()
             updater(X.shape[0])
-            metric.add(float(l.sum()), accuracy(y_hat, y), y.numel())
+        metric.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]
 
@@ -1415,8 +1413,8 @@ def train_concise_ch11(trainer_fn, hyperparams, data_iter, num_epochs=4):
 
     optimizer = trainer_fn(net.parameters(), **hyperparams)
 
-    loss = nn.MSELoss()
-    # L2Loss=1/2*MSELoss
+    # 注意：MSELoss计算平方误差时不带系数1/2
+    loss = nn.MSELoss(reduction='none')
     animator = d2l.Animator(xlabel='epoch', ylabel='loss',
                             xlim=[0, num_epochs], ylim=[0.22, 0.35])
     n, timer = 0, d2l.Timer()
@@ -1425,14 +1423,14 @@ def train_concise_ch11(trainer_fn, hyperparams, data_iter, num_epochs=4):
             optimizer.zero_grad()
             out = net(X)
             y = y.reshape(out.shape)
-            l = loss(out, y)/2
-            l.backward()
+            l = loss(out, y)
+            l.mean().backward()
             optimizer.step()
             n += X.shape[0]
             if n % 200 == 0:
                 timer.stop()
                 animator.add(n/X.shape[0]/len(data_iter),
-                             (d2l.evaluate_loss(net, data_iter, loss)/2,))
+                             (d2l.evaluate_loss(net, data_iter, loss),))
                 timer.start()
     print(f'loss: {animator.Y[0][-1]:.3f}, {timer.avg():.3f} sec/epoch')