Beginning Anomaly Detection Using

n means a dilation factor of 

n. A tuple of two integers allows you to specify (vertical_dilation, 

horizontal_dilation). Default = 1.

• 

return_indices: If set to True, it will return the indices of the max 

values along with the outputs. Default = False.

• 

ceil_mode: If set to True, it will use ceil instead of floor to compute 

the output shape. This comes into play because of the dimensionality 

reduction involved (a kernel size of 

n will reduce dimensionality by a 

factor of 

n).

 ZeroPadding2D

torch.nn.ZeroPad2d()

Depending on the input, it pads the input sequence with a row and columns of 

zeroes at the top, left, right, and bottom of the image tensor.

appendix B   intro to pytorch

---

382

Here is the parameter:

• 

padding: An integer or a tuple of four integers. The integer tells it 

to add 

n rows of zeroes on the top and bottom of the image tensor, 

and 

n columns of zeroes. The tuple of four integers is formatted as 

(padding_left, padding_right, padding_top, padding_bottom), so 

you can customize even more how you want the layer to add rows or 

columns of zeroes.

 Dropout

torch.nn.Dropout()

What the dropout layer does in PyTorch is take the input and randomly zeroes the 

elements according to some probability 

p using samples from a Bernoulli distribution. 

This process is random, so with every forward pass through the model, different 

elements will be chosen to be zeroed. This process helps with regularization of layer 

outputs and helps combat overfitting.

Here are the parameters:

• 

p: The probability of an element to be zeroed. Default = 0.5

• 

inplace: If set to True, it will perform the operation in place.  

Default = False.

You can define this as a layer within the model itself, or apply dropout in the forward 

function like so:

torch.nn.functional.Dropout(input, p = 0.5, training=False, inplace=False)