Скачать или смотреть L07.4: Writing your own Training and Evaluation Loops

In this video I take a closer look at the `fit()` training loop that Keras implements. As a reminder, the contents of a typical training loop look like

1. Run the forward pass inside a gradient tape to obtain loss for a batch
2. Retrieve gradients of the loss with respect to the model's weights
3. Update the model's weights

We can implement these steps by hand easily enough with the help of the tensorflow tape and completely implement a training loop from scratch. But I also look at a probably more intermediate method in this video, subclassing the `Model` class again to override the `train_step()` method if we need fine control of the training loop but want to reuse all of the other Keras framework such as metrics and loss functions.

Resources:

Textbook: Chollet (2022). "Deep Learning with Python (2ed)". Manning.
https://www.amazon.com/dp/1617296864/...

CSci 560 Class Repository: https://github.com/csci560-nndl/nndl
Contains video slides and iPython notebooks for this course.



00:00 Introduction
00:40 Writing your own Training and Evaluation loops
04:59 An example from-scratch training loop and evaluation loop implementation
08:40 Making it fast by using @tf.function decorator
10:30 Subclass `Model` and leverage `fit()` with a custom training loop
13:30 Summary