A Little Tips on Constructing Neuronal Network Based on Pytorch（1）

A Little Tips on Constructing Neuronal Network Based on Pytorch（1）

In python coding, we can divide the whole code into some parts, difference of every part of code with different functions.

Using Manual Experiment of Fully Connected Neuronal Network based on MNIST as an example, in this task, we have several subtasks:

• super-parameters setting
• data preparation
• constructing neuronal network
• train the neuronal network
• verify the neuronal network

By analyzing the requirements of each subtasks, we can define several modules in python project:

• configuration.py
• datasets.py
• FNN_Model.py
• train.py
• evalue.py
• main.py

The project folder can be set as follows:

PS:In the project folder we can also add models subfolder for saving trained model parameters or data subfolder for managing data files.