RuntimeError: Calculated padded input size per channel: (3). Kernel size: (20). Kernel size can't be greater than actual input size

nlp
1

Hi,

I’m so new at Pytorch, and I’m trying to convert my Keras model into Pytorch one. In Keras, I can use the kernel_size = 20 with the same data and I worked well. However, when I define as 20 with the same data and architectire in Pytorch, I got the following error. I can’t really understand the main reason for this error since It was working well in Keras.
I’d be appreciated If you could make me understand the issue and solution.

RuntimeError                              Traceback (most recent call last)
<ipython-input-25-74505e66d6bf> in <module>
     17 
     18         # Forward pass
---> 19         outputs = model(dataX.float())
     20         loss = criterion(outputs, labels)
     21 

~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
    887             result = self._slow_forward(*input, **kwargs)
    888         else:
--> 889             result = self.forward(*input, **kwargs)
    890         for hook in itertools.chain(
    891                 _global_forward_hooks.values(),

<ipython-input-19-6d0943e3b0de> in forward(self, x)
     28         x = self.maxpool1(x)
     29         print("x", x.size())
---> 30         x = self.cnn3(x)
     31         print("cnn3", x.size())
     32         x = self.act3(x)

~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
    887             result = self._slow_forward(*input, **kwargs)
    888         else:
--> 889             result = self.forward(*input, **kwargs)
    890         for hook in itertools.chain(
    891                 _global_forward_hooks.values(),

~\Anaconda3\lib\site-packages\torch\nn\modules\conv.py in forward(self, input)
    261 
    262     def forward(self, input: Tensor) -> Tensor:
--> 263         return self._conv_forward(input, self.weight, self.bias)
    264 
    265 

~\Anaconda3\lib\site-packages\torch\nn\modules\conv.py in _conv_forward(self, input, weight, bias)
    257                             weight, bias, self.stride,
    258                             _single(0), self.dilation, self.groups)
--> 259         return F.conv1d(input, weight, bias, self.stride,
    260                         self.padding, self.dilation, self.groups)
    261 

RuntimeError: Calculated padded input size per channel: (3). Kernel size: (20). Kernel size can't be greater than actual input size```
2

Could you post the model definition in Keras as well as PyTorch?
I guess you might use different conv settings, e.g. using padding='same' in Keras while not using any padding in PyTorch, which would yield different output activation shapes.

3

I’m sorry, I can’t post the model definition, unfortunately (my company policies). I used padding default which is same, but I also defined my padding=1 in everywhere which is the equivalent of the Keras model I wrote. As far as I know, Keras is padding with the required amount of zeros for the filter, yet this is also equivalent to apply convolution on a bunch of zeros. Therefore, I updated the model parameters as Pytorch asks. So, problem solved. Thanks anyway.