Beginning Anomaly Detection Using

 Sequential vs. ModuleList

Similar to Keras, PyTorch has a couple different ways to define the model.

Sequentially, as in Figure 

B-16

Figure B-16.  A sequential model in PyTorch

This is similar to the sequential model in Keras, where you add layers one at a time 

and in order.

ModuleList, as in Figure 

B-17

appendix B   intro to pytorch

---

377

Figure B-17.  A model in PyTorch defined in a ModuleList format

This is similar to the functional model that you can build in Keras. This is a more 

customizable way to build your model, and allows you much more flexibility in how you 

want to build it too.

 Layers

We’ve covered how to build the models, so let’s look at examples of some common layers 

you can build.

 Conv1d

torch.nn.Conv1d()

Check out Chapter 

7

 for a detailed explanation on how one-dimensional 

convolutions work.

appendix B   intro to pytorch

---

378

This layer is a one-dimensional (or temporal) convolutional layer. It basically passes 

a filter over the one-dimensional input and multiplies the values element-wise to create 

the output feature map.

These are the parameters that the function takes:

• 

in_channels: The dimensionality of the input space; the number of 

input nodes.

• 

out_channels: The dimensionality of the output space; the number 

of output nodes.

•