[PaddlePaddle] fix d2l.py bugs (d2l-ai#1181)

* update d2l

* Fix the bugs of d2l

* remove multi gpus paddle tab

Co-authored-by: cheungdaven <cheungdaven@gmail.com>

chapter_computational-performance/multiple-gpus-concise.md

@@ -317,7 +317,7 @@ train(num_gpus=2, batch_size=512, lr=0.2)
 ```
 
 ```{.python .input}
-#@tab pytorch, paddle
+#@tab pytorch
 train(net, num_gpus=2, batch_size=512, lr=0.2)
 ```
 

chapter_computational-performance/multiple-gpus.md

@@ -544,7 +544,7 @@ train(num_gpus=1, batch_size=256, lr=0.2)
 尽管如此，让我们看看Fashion-MNIST数据集上会发生什么。
 
 ```{.python .input}
-#@tab all
+#@tab mxnet, pytorch
 train(num_gpus=2, batch_size=256, lr=0.2)
 ```
 

chapter_computer-vision/semantic-segmentation-and-dataset.md

@@ -112,6 +112,7 @@ train_features, train_labels = read_voc_images(voc_dir, True)
 #@tab paddle
 #@save
 def read_voc_images(voc_dir, is_train=True):
+    """Defined in :numref:`sec_semantic_segmentation`"""
     """读取所有VOC图像并标注
     Defined in :numref:`sec_semantic_segmentation`"""
     txt_fname = os.path.join(voc_dir, 'ImageSets', 'Segmentation',
@@ -120,11 +121,11 @@ def read_voc_images(voc_dir, is_train=True):
         images = f.read().split()
     features, labels = [], []
     for i, fname in enumerate(images):
-        features.append(paddle.to_tensor(paddle.vision.image.image_load(os.path.join(
-            voc_dir, 'JPEGImages', f'{fname}.jpg'), backend='cv2')[..., ::-1], dtype=paddle.float32).transpose(
+        features.append(paddle.vision.image.image_load(os.path.join(
+            voc_dir, 'JPEGImages', f'{fname}.jpg'), backend='cv2')[..., ::-1].transpose(
             [2, 0, 1]))
-        labels.append(paddle.to_tensor(paddle.vision.image.image_load(os.path.join(
-            voc_dir, 'SegmentationClass', f'{fname}.png'), backend='cv2')[..., ::-1], dtype=paddle.float32).transpose(
+        labels.append(paddle.vision.image.image_load(os.path.join(
+            voc_dir, 'SegmentationClass', f'{fname}.png'), backend='cv2')[..., ::-1].transpose(
             [2, 0, 1]))
     return features, labels
 
@@ -152,7 +153,7 @@ d2l.show_images(imgs, 2, n);
 #@tab paddle
 n = 5
 imgs = train_features[0:n] + train_labels[0:n]
-imgs = [img.transpose([1, 2, 0]).astype(paddle.uint8) for img in imgs]
+imgs = [img.transpose([1, 2, 0]) for img in imgs]
 d2l.show_images(imgs, 2, n);
 ```
 
@@ -230,7 +231,7 @@ def voc_colormap2label():
 
 def voc_label_indices(colormap, colormap2label):
     """将VOC标签中的RGB值映射到它们的类别索引"""
-    colormap = colormap.transpose([1, 2, 0]).astype('int32')#[281,500,3]
+    colormap = colormap.transpose([1, 2, 0]).astype('int32')
     idx = ((colormap[:, :, 0] * 256 + colormap[:, :, 1]) * 256
            + colormap[:, :, 2])
     return colormap2label[idx]
@@ -306,9 +307,9 @@ d2l.show_images(imgs[::2] + imgs[1::2], 2, n);
 #@tab paddle
 imgs = []
 for _ in range(n):
-    imgs += voc_rand_crop(train_features[0], train_labels[0], 200, 300)
-
-imgs = [img.transpose([1, 2, 0]).astype(paddle.uint8) for img in imgs]
+    imgs += voc_rand_crop(train_features[0].transpose([1, 2, 0]), train_labels[0].transpose([1, 2, 0]), 200, 300)
+    
+imgs = [img for img in imgs]
 d2l.show_images(imgs[::2] + imgs[1::2], 2, n);
 ```
 
@@ -390,7 +391,8 @@ class VOCSegDataset(torch.utils.data.Dataset):
 #@tab paddle
 #@save
 class VOCSegDataset(paddle.io.Dataset):
-    """一个用于加载VOC数据集的自定义数据集"""
+    """一个用于加载VOC数据集的自定义数据集
+    Defined in :numref:`sec_semantic_segmentation`"""
 
     def __init__(self, is_train, crop_size, voc_dir):
         self.transform = paddle.vision.transforms.Normalize(
@@ -404,17 +406,18 @@ class VOCSegDataset(paddle.io.Dataset):
         print('read ' + str(len(self.features)) + ' examples')
 
     def normalize_image(self, img):
-        return self.transform(img.astype(paddle.float32) / 255)
+        return self.transform(img.astype("float32") / 255)
 
     def filter(self, imgs):
         return [img for img in imgs if (
             img.shape[1] >= self.crop_size[0] and
             img.shape[2] >= self.crop_size[1])]
 
     def __getitem__(self, idx):
-        feature, label = voc_rand_crop(self.features[idx], self.labels[idx],
+        feature = paddle.to_tensor(self.features[idx],dtype='float32')
+        label = paddle.to_tensor(self.labels[idx],dtype='float32')
+        feature, label = voc_rand_crop(feature,label,
                                        *self.crop_size)
-        
         return (feature, voc_label_indices(label, self.colormap2label))
 
     def __len__(self):

chapter_convolutional-neural-networks/lenet.md

@@ -209,6 +209,7 @@ def evaluate_accuracy_gpu(net, data_iter, device=None):     #@save
         net.eval()  # 设置为评估模式
         if not device:
             device = next(iter(net.parameters())).place
+    paddle.set_device("gpu:{}".format(str(device)[-2]))
     # 正确预测的数量，总预测的数量
     metric = d2l.Accumulator(2)
     with paddle.no_grad():

chapter_linear-networks/image-classification-dataset.md

@@ -253,7 +253,7 @@ batch_size = 256
 
 def get_dataloader_workers():  #@save
     """使用4个进程来读取数据"""
-    return 0 if not sys.platform.startswith('linux') else 4
+    return 0
 
 train_iter = paddle.io.DataLoader(dataset=mnist_train,
                                   batch_size=batch_size,

chapter_preface/index.md

@@ -134,6 +134,7 @@ import paddle
 from paddle import nn
 from paddle.nn import functional as F
 from paddle.vision import transforms
+import paddle.vision as paddlevision
 from PIL import Image
 ```
 

chapter_recurrent-modern/seq2seq.md

@@ -76,8 +76,6 @@ from d2l import paddle as d2l
 import math
 import paddle
 from paddle import nn
-import warnings
-warnings.filterwarnings("ignore")
 ```
 
 ## 编码器
@@ -828,7 +826,7 @@ def train_seq2seq(net, data_iter, lr, num_epochs, tgt_vocab, device):
             X, X_valid_len, Y, Y_valid_len = [paddle.to_tensor(x, place=device) for x in batch]
             bos = paddle.to_tensor([tgt_vocab['<bos>']] * Y.shape[0]).reshape([-1, 1])
             dec_input = paddle.concat([bos, Y[:, :-1]], 1)  # 强制教学
-            Y_hat, _ = net(X, dec_input, X_valid_len)
+            Y_hat, _ = net(X, dec_input, X_valid_len.squeeze())
             l = loss(Y_hat, Y, Y_valid_len.squeeze())
             l.backward()	# 损失函数的标量进行“反向传播”
             d2l.grad_clipping(net, 1)

d2l/mxnet.py

@@ -801,7 +801,7 @@ def show_list_len_pair_hist(legend, xlabel, ylabel, xlist, ylist):
     for patch in patches[1].patches:
         patch.set_hatch('/')
     d2l.plt.legend(legend)
-    
+
 def truncate_pad(line, num_steps, padding_token):
     """截断或填充文本序列