This is a demo code made for Prof.CHEN Yingcong's research group students. It serves as a training callback template for students to collaborate. For demo video recording, please see Notion page for details.
Part of the code is taken from Fastai 2019 part 2 course, where I was invited as an international fellow.
- Data, MNIST original, data are length 784 vectors
- Model, a tiny MLP with 2 linear layer with a relu between them
- Loss, cross entropy
Please scroll down the demo notebook, where you will see implemented example callbacks, they are SetUpCallback, LoopControlCallback, LogCallback and more...
To understand the notebook:
- The Learner_Nocallback class shows a basic training loop with no cbs
- The NewLeaner class shows a training loop with callback and event (exceptions) controlled loop