Getting started

For this tutorial, we will only focuses in the implementation of the fully connected part of Jeremy Fix’s tutorial. In this page, we gonna see how generate the project and the role of each file.

Project generation

First, we need to generate an empty JAW project.

python3 src/jaw/JAWProject.py

Enter the name of your project and where save it.

Warning

Use an absolute path for your project filepath.

You will have this result:

yvregon@nostradamus:~/JAW$ python3 src/jaw/JAWProject.py
Enter a project name : tutorial
Enter a root folder (default : /home/yvregon/JAW) :
New JAW project generated.

And the following structure:

A picture of a folder inside VSCode, with the project structure. — Our project structure.

File tree

For now, you will have a tree file like this :

project_name
├── data_preprocessing
│   └── data_utils.py
├── losses
│   └── loss.py
├── models
│   └── CustomModel.py
├── training
│   ├── evaluation.py
│   └── train.py
└── project_name_trainer.py

Let’s have a closer look to this folder structure:

project_name is the root of our project.
data_preprocessing is where we put our code for everything related to the data preprocessing, such as dataset transformer or splitting method.
losses and models are the folder where put our custom network models and loss classes.
The training folder contain our functions that handle training and evaluation loops.
project_name_trainer.py is our “main” script: it’s here where we define our training process and launch options.

Tip

The project generation script produce automatic reStructuredText docstring, but you can also generate the same Sphinx documentation using a more readable docstring with napoleon extention (used by numpy for instance).