As a bonus, get the code to use feature extraction anywhere

That is the last a part of my series of nature-inspired articles. Earlier, I had talked about algorithms inspired by genetics, swarm, bees, and ants. Today, I’ll speak about wolves.

When a journal paper has a citation count spanning 5 figures, you recognize there’s some serious business happening. Grey Wolf Optimizer [1] (GWO) is one such example.

Like Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), and Ant Colony Optimization (ACO), GWO can be a meta-heuristic. Although there’s no mathematical guarantees to the answer, it really works well in practice and doesn’t require any analytical knowledge of the underlying problem. This permits us to question from a ‘blackbox’, and easily make use to the observed results to refine our solution.

As mentioned in my ACO article, all these ultimately relate back to the basic concept of explore-exploit trade-off. Why, then, are there so many alternative meta-heuristics?

Firstly, it’s because researchers need to publish papers. A great a part of their job entails exploring things from different angles and sharing the ways through which their findings bring about advantages over existing approaches. (Or as some would say, publishing papers to justify their salaries and seek promotions. But let’s not get there.)

Secondly, it’s attributable to the ‘No Free Lunch’ theorem [2] which the authors of GWO themselves talked about. While that theorem was specifically saying there’s no free lunch for optimization algorithms, I believe it’s fair to say that the identical is true for Data Science normally. There isn’t a single ultimate one-size-fits-all solution, and we regularly need to try different approaches to see what works.

Subsequently, let’s proceed so as to add yet one more meta-heuristic to our toolbox. Since it never hurts to have one other tool which could come in useful in the future.

First, let’s consider a straightforward classification problem on images. A clever approach is to make use of pre-trained deep neural networks as feature extractors, to convert…