How To Define A Sequential Neural Network Container In PyTorch

Use PyTorch's nn.Sequential and add_module operations to define a sequential neural network container

Type: FREE   By: Tylan O'Flynn, AIWorkbox.com Instructor Tylan O'Flynn   Duration: 1:38   Technologies: PyTorch, Python

Page Sections: Video  |  Code  |  Transcript


< > Code:

You must be a Member to view code

Access all courses and lessons, gain confidence and expertise, and learn how things work and how to use them.

    or   Log In


Transcript:

Once our data has been imported and pre-processed, the next step is to build the neural network that we'll be training and testing using the data.

Though our ultimate goal is to use a more complex model to process the data, such as a residual neural network, we will start with a simple convolutional neural network or CNN.


First, we will import the necessary packages.


For the neural network, we will need torch

import torch


torch.nn as nn

import torch.nn as nn


and from the collections package, we will also want to import OrderedDict as this will be necessary later in the example.

from collections import OrderedDict


For the convolutional neural network, first we will need to define a container.

Containers can be defined as sequential, module list, module dictionary, parameter list, or parameter dictionary.

The sequential, module list, and module dictionary containers are the highest level containers and can be thought of as neural networks with no layers added in.


The sequential container can be defined as follows:

model = nn.Sequential(OrderedDict([
          ('conv1', nn.Conv2d(1,20,5)),
          ('relu1', nn.ReLU()),
          ('conv2', nn.Conv2d(20,64,5)),
          ('relu2', nn.ReLU())
        ]))

Model, in this case our neural network, equals nn.Sequential with an OrderedDict of various layers as an argument.


I can define the container with layers as arguments as I have done here with an OrderedDict, but I can also add layers to the end.

This is achieved using add_module.

model.add_module('conv3', nn.Conv2d(64, 64, 5))

The arguments for add_module are a name for the layer and the nature of the layer, in this case 2d convolutional layer.


Though it is not necessary, it is good practice to use OrderedDict when creating a sequential model as well as passing a name string as a parameter when using add_module as this one will make your model much, much easier to inspect and debug.



Back to deep learning tutorial lesson list