CNN Pytorch Error : Input type (torch.cuda.ByteTensor) and weight type (torch.cuda.FloatTensor) should be the same

huangapple
go评论120阅读模式
英文:

CNN Pytorch Error : Input type (torch.cuda.ByteTensor) and weight type (torch.cuda.FloatTensor) should be the same

问题

我收到了以下错误信息：

输入类型（torch.cuda.ByteTensor）和权重类型（torch.cuda.FloatTensor）应该相同

以下是我的代码：

  1. device = torch.device('cuda:0')
  3. trainData = torchvision.datasets.FashionMNIST('/content/', train=True, transform=None, target_transform=None, download=True)
  4. testData = torchvision.datasets.FashionMNIST('/content/', train=False, transform=None, target_transform=None, download=True)
  6. class Net(nn.Module):
  7.   def __init__(self):
  8.     super().__init__()
  10.     '''
  11.     网络结构：
  13.     输入 > 
  14.     (1)Conv2D > (2)MaxPool2D > 
  15.     (3)Conv2D > (4)MaxPool2D > 
  16.     (5)Conv2D > (6)MaxPool2D > 
  17.     (7)Linear > (8)LinearOut
  19.     '''
  21.     # 创建卷积层
  22.     self.conv1 = nn.Conv2d(in_channels=CHANNELS, out_channels=32, kernel_size=3)
  23.     self.conv2 = nn.Conv2d(in_channels=32, out_channels=64, kernel_size=3)
  24.     self.conv3 = nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3)
  26.     self.flatten = None
  27.     # 创建一个随机的虚拟样本以获取平铺的维度
  28.     x = torch.randn(CHANNELS, DIM, DIM).view(-1, CHANNELS, DIM, DIM)
  29.     x = self.convs(x)
  31.     # 创建线性层
  32.     self.fc1 = nn.Linear(self.flatten, 512)
  33.     self.fc2 = nn.Linear(512, CLASSES)
  35.   def convs(self, x):
  37.     # 创建最大池化层
  38.     x = F.max_pool2d(F.relu(self.conv1(x)), kernel_size=(2, 2))
  39.     x = F.max_pool2d(F.relu(self.conv2(x)), kernel_size=(2, 2))
  40.     x = F.max_pool2d(F.relu(self.conv3(x)), kernel_size=(2, 2))
  42.     if not self.flatten:
  43.       self.flatten = x[0].shape[0] * x[0].shape[1] * x[0].shape[2]
  44.     return x
  46.   # 前向传播
  47.   def forward(self, x):
  48.     x = self.convs(x)
  49.     x = x.view(-1, self.flatten)
  50.     sm = F.relu(self.fc1(x))
  51.     x = F.softmax(self.fc2(sm), dim=1)
  52.     return x, sm
  54. x_train, y_train = training_set
  55. x_train, y_train = x_train.to(device), y_train.to(device)
  56. optimizer = optim.Adam(net.parameters(), lr=LEARNING_RATE)
  57. loss_func = nn.MSELoss()
  58. loss_log = []
  60. for epoch in range(EPOCHS):
  61.   for i in tqdm(range(0, len(x_train), BATCH_SIZE)):
  62.     x_batch = x_train[i:i+BATCH_SIZE].view(-1, CHANNELS, DIM, DIM).to(device)
  63.     y_batch = y_train[i:i+BATCH_SIZE].to(device)
  65.     net.zero_grad()
  66.     output, sm = net(x_batch)
  67.     loss = loss_func(output, y_batch.float())
  68.     loss.backward()
  69.     optimizer.step()
  70.   loss_log.append(loss)
  71.   # print(f"Epoch : {epoch} || Loss : {loss}")
  73. return loss_log
  75. train_set = (trainData.train_data, trainData.train_labels)
  76. test_set = (testData.test_data, testData.test_labels)
  78. EPOCHS = 5
  79. LEARNING_RATE = 0.001
  80. BATCH_SIZE = 32
  82. net = Net().to(device)
  84. loss_log = train(net, train_set, EPOCHS, LEARNING_RATE, BATCH_SIZE)

这是我收到的错误信息：

  1. RuntimeError                              Traceback (most recent call last)
  2. <ipython-input-8-0db1a1b4e37d> in <module>()
  3.       5 net = Net().to(device)
  4.       6 
  5. ----> 7 loss_log = train(net, train_set, EPOCHS, LEARNING_RATE, BATCH_SIZE)
  7. 6 frames
  8. <ipython-input-6-7de4a78e3736> in train(net, training_set, EPOCHS, LEARNING_RATE, BATCH_SIZE)
  9.      13 
  10.      14         net.zero_grad()
  11. ---> 15         output, sm = net(x_batch)
  12.      16         loss = loss_func(output, y_batch.float())
  13.      17         loss.backward()
  15. /usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
  16.     539             result = self._slow_forward(*input, **kwargs)
  17.     540         else:
  18. --> 541             result = self.forward(*input, **kwargs)
  19.     542         for hook in thismodel.forward_features(x)
  20.     543             hook_result = hook(thismodel, input, result)
  22. <ipython-input-5-4fddc427892a> in forward(self, x)
  23.      41   # 前向传播
  24.      42   def forward(self, x):
  25. ---> 43     x = self.convs(x)
  26.      44     x = x.view(-1, self.flatten)
  27.      45     sm = F.relu(self.fc1(x))
  29. <ipython-input-5-4fddc427892a> in convs(self, x)
  30.      31 
  31.      32     # 创建最大池化层
  32. ---> 33     x = F.max_pool2d(F.relu(self.conv1(x)), kernel_size=(2, 2))
  33.      34     x = F.max_pool2d(F.relu(self.conv2(x)), kernel_size=(2, 2))
  34.      35     x = F.max_pool2d(F.relu(self.conv3(x)), kernel_size=(2, 2))
  36. /usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
  37.     539             result = thismodel.features
  38.     540         else:
  39. --> 541             result = thismodel.forward(*input, **kwargs)
  40.     542         for hook in thismodel.forward_features(x)
  41.     543             hook_result = hook(thismodel, input, result)
  43. <ipython-input-5-4fddc427892a> in forward(self, x)
  44.      41   # 前向传播
  45.      42   def forward(self, x):
  46. ---> 43     x = self.convs(x)
  47.      44     x = x.view(-1, self.flatten)
  48.      45     sm = F.relu(self.fc1(x))
  50. <ipython-input-5-4fddc427892a> in convs(self, x)
  51.      31 
  52.      32     # 创建最大池化层
  53. ---> 33     x = F.max_pool2d(F.relu(self.conv1(x)), kernel_size=(2, 2))
  54.      34     x = F.max_pool2d(F.relu(self.conv2(x)), kernel_size=(2
  56. <details>
  57. <summary>英文:</summary>
  59. I&#39;m receiving the error,
  60. &gt; Input type (torch.cuda.ByteTensor) and weight type (torch.cuda.FloatTensor) should be the same
  62. Following is my code,
  64.     device    = torch.device(&#39;cuda:0&#39;)
  65.     
  66.     trainData = torchvision.datasets.FashionMNIST(&#39;/content/&#39;, train=True, transform=None, target_transform=None, download=True)
  67.     testData  = torchvision.datasets.FashionMNIST(&#39;/content/&#39;, train=False, transform=None, target_transform=None, download=True)
  68.     
  69.     class Net(nn.Module):
  70.       def __init__(self):
  71.         super().__init__()
  72.     
  73.         &#39;&#39;&#39;
  74.         Network Structure:
  75.     
  76.         input &gt; 
  77.         (1)Conv2D &gt; (2)MaxPool2D &gt; 
  78.         (3)Conv2D &gt; (4)MaxPool2D &gt; 
  79.         (5)Conv2D &gt; (6)MaxPool2D &gt; 
  80.         (7)Linear &gt; (8)LinearOut
  81.     
  82.         &#39;&#39;&#39;
  83.     
  84.         # Creating the convulutional Layers
  85.         self.conv1 = nn.Conv2d(in_channels=CHANNELS, out_channels=32, kernel_size=3)
  86.         self.conv2 = nn.Conv2d(in_channels=32, out_channels=64, kernel_size=3)
  87.         self.conv3 = nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3)
  88.     
  89.         self.flatten = None
  90.         # Creating a Random dummy sample to get the Flattened Dimensions
  91.         x = torch.randn(CHANNELS, DIM, DIM).view(-1, CHANNELS, DIM, DIM)
  92.         x = self.convs(x)
  93.     
  94.         # Creating the Linear Layers
  95.         self.fc1   = nn.Linear(self.flatten, 512)
  96.         self.fc2   = nn.Linear(512, CLASSES)
  97.     
  98.       def convs(self, x):
  99.     
  100.         # Creating the MaxPooling Layers
  101.         x = F.max_pool2d(F.relu(self.conv1(x)), kernel_size=(2, 2))
  102.         x = F.max_pool2d(F.relu(self.conv2(x)), kernel_size=(2, 2))
  103.         x = F.max_pool2d(F.relu(self.conv3(x)), kernel_size=(2, 2))
  104.     
  105.         if not self.flatten:
  106.           self.flatten = x[0].shape[0] * x[0].shape[1] * x[0].shape[2]
  107.         return x
  108.     
  109.       # FORWARD PASS
  110.       def forward(self, x):
  111.         x = self.convs(x)
  112.         x = x.view(-1, self.flatten)
  113.         sm = F.relu(self.fc1(x))
  114.         x = F.softmax(self.fc2(sm), dim=1)
  115.         return x, sm
  116.     
  117.     
  118.       x_train, y_train = training_set
  119.       x_train, y_train = x_train.to(device), y_train.to(device)
  120.       optimizer = optim.Adam(net.parameters(), lr=LEARNING_RATE)
  121.       loss_func = nn.MSELoss()
  122.       loss_log  = []
  123.     
  124.       for epoch in range(EPOCHS):
  125.         for i in tqdm(range(0, len(x_train), BATCH_SIZE)):
  126.             x_batch = x_train[i:i+BATCH_SIZE].view(-1, CHANNELS, DIM, DIM).to(device)
  127.             y_batch = y_train[i:i+BATCH_SIZE].to(device)
  128.     
  129.             net.zero_grad()
  130.             output, sm = net(x_batch)
  131.             loss = loss_func(output, y_batch.float())
  132.             loss.backward()
  133.             optimizer.step()
  134.         loss_log.append(loss)
  135.         # print(f&quot;Epoch : {epoch} || Loss : {loss}&quot;)
  136.     
  137.       return loss_log
  138.     
  139.     
  140.     train_set = (trainData.train_data, trainData.train_labels)
  141.     test_set  = (testData.test_data, testData.test_labels)
  142.     
  143.     EPOCHS        = 5
  144.     LEARNING_RATE = 0.001
  145.     BATCH_SIZE    = 32
  146.     
  147.     net = Net().to(device)
  148.     
  149.     loss_log = train(net, train_set, EPOCHS, LEARNING_RATE, BATCH_SIZE)
  151. **And this is the Error that I&#39;m getting,**
  153.     RuntimeError                              Traceback (most recent call last)
  154.     &lt;ipython-input-8-0db1a1b4e37d&gt; in &lt;module&gt;()
  155.           5 net = Net().to(device)
  156.           6 
  157.     ----&gt; 7 loss_log = train(net, train_set, EPOCHS, LEARNING_RATE, BATCH_SIZE)
  158.     
  159.     6 frames
  160.     &lt;ipython-input-6-7de4a78e3736&gt; in train(net, training_set, EPOCHS, LEARNING_RATE, BATCH_SIZE)
  161.          13 
  162.          14         net.zero_grad()
  163.     ---&gt; 15         output, sm = net(x_batch)
  164.          16         loss = loss_func(output, y_batch.float())
  165.          17         loss.backward()
  166.     
  167.     /usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
  168.         539             result = self._slow_forward(*input, **kwargs)
  169.         540         else:
  170.     --&gt; 541             result = self.forward(*input, **kwargs)
  171.         542         for hook in self._forward_hooks.values():
  172.         543             hook_result = hook(self, input, result)
  173.     
  174.     &lt;ipython-input-5-4fddc427892a&gt; in forward(self, x)
  175.          41   # FORWARD PASS
  176.          42   def forward(self, x):
  177.     ---&gt; 43     x = self.convs(x)
  178.          44     x = x.view(-1, self.flatten)
  179.          45     sm = F.relu(self.fc1(x))
  180.     
  181.     &lt;ipython-input-5-4fddc427892a&gt; in convs(self, x)
  182.          31 
  183.          32     # Creating the MaxPooling Layers
  184.     ---&gt; 33     x = F.max_pool2d(F.relu(self.conv1(x)), kernel_size=(2, 2))
  185.          34     x = F.max_pool2d(F.relu(self.conv2(x)), kernel_size=(2, 2))
  186.          35     x = F.max_pool2d(F.relu(self.conv3(x)), kernel_size=(2, 2))
  187.     
  188.     /usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
  189.         539             result = self._slow_forward(*input, **kwargs)
  190.         540         else:
  191.     --&gt; 541             result = self.forward(*input, **kwargs)
  192.         542         for hook in self._forward_hooks.values():
  193.         543             hook_result = hook(self, input, result)
  194.     
  195.     /usr/local/lib/python3.6/dist-packages/torch/nn/modules/conv.py in forward(self, input)
  196.         343 
  197.         344     def forward(self, input):
  198.     --&gt; 345         return self.conv2d_forward(input, self.weight)
  199.         346 
  200.         347 class Conv3d(_ConvNd):
  201.     
  202.     /usr/local/lib/python3.6/dist-packages/torch/nn/modules/conv.py in conv2d_forward(self, input, weight)
  203.         340                             _pair(0), self.dilation, self.groups)
  204.         341         return F.conv2d(input, weight, self.bias, self.stride,
  205.     --&gt; 342                         self.padding, self.dilation, self.groups)
  206.         343 
  207.         344     def forward(self, input):
  208.     
  209.     RuntimeError: Input type (torch.cuda.ByteTensor) and weight type (torch.cuda.FloatTensor) should be the same
  211. I double-checked that my Neural Net and my Inputs both are in GPU. I&#39;m still getting this error and I don&#39;t understand why!
  213. Somebody, please help me to get out of this error.
  216. </details>
  219. # 答案1
  220. **得分**: 23
  222. 将输入的 `x_batch` 转换为浮点型。在将其传递到模型之前使用 `x_batch = x_batch.float()`。
  224. <details>
  225. <summary>英文:</summary>
  227. Cast your input `x_batch` to float. Use `x_batch = x_batch.float()` before you pass it through your model. 
  229. </details>

huangapple
  • 本文由 发表于 2020年1月4日 01:15:36
  • 转载请务必保留本文链接：https://go.coder-hub.com/59582663.html
  • conv-neural-network
  • machine-learning
  • python
  • pytorch
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定