This is how the ML practitioner does their work. Start small and do all the experiments slowly scale up the model.

In this module we'll be using 101 food classes which is all of the classes but with only 10% of that data.

This is more like a iterative experimentation,

• Unfreeze more layers gradually.
• Decrease the learning rate
• Compile the model and Fit the model
• Repeat —> Till the results turns out to be good

Fine-tuning won't always necessarily improve the results, it's another experiment we're running.

While making predictions we gotta make sure the data we passed in should be same of the data it had trained on.