🍳This project aims at providing a rich function environment for Machine Learning with PyTorch.🍳
🍻🍻Contributions welcome!
In this framework, the Dataset, Net, Action are separated totally. They are powered by args.
The basic framework structure illustrates as below.
Dataset: You can define all your datasets here, and use get_dataloader function to get its PyTorch dataloader.
Net: This module includes all the models your project needed. And to reuse some sub-models, you can define them in net_parts.
Action: The Action module occupies a very important position as you can code any action you want to record your network results, such as model graph,accuracy, confusion matrix, auc and loss etc..
Agent: It is the spokesman of the above three modules, because model train/eval with dataset and using action to record the performance. So, in this part, you can assign dataloader, model and action by get_dataloader, get_action and get_net function respectively.
Config: This part mainly used to config some directories which can be dirs to store results or dirs referring to datasets. It contains the environment which remains unchanged. It can be useful when you debug locally and deploy remotely.
Tool: It is a utils combination mainly used to generate parser. But you can put any mess code here.
Main: It is the core of the framework as it provides args to other modules. We use args to control the whole work.
Logger: This tool is a colorful progressbar logger. It can be used simply as a logger. During training, you can activate its progressbar function, so you can view the progress of your program.
Some screenshot shows as below:
