Spot-checking extractors and classifiers

Often, when we are tackling a new project, we are victims of the Paradox of Choice: we don't know where or how to start due to the presence of so many options to choose from. Which feature extractor is the best? What's the most performant model we can train? How should we pre-process our data?

In this recipe, we will implement a framework that will automatically spot-check feature extractors and classifiers. The goal is not to get the best possible model right away, but to narrow down our options so that we can focus on the most promising ones at a later stage.

Getting ready

First, we must install Pillow and tqdm:

$> pip install Pillow tqdm

We'll use a dataset called 17 Category Flower Dataset, available here: http://www.robots.ox.ac.uk/~vgg/data/flowers/17. However, a curated version, organized into subfolders per class, can be downloaded here: https://github.com/PacktPublishing/Tensorflow-2.0-Computer-Vision...