A Cutesy/Silly Introduction to Machine Learning for Tech and Non Tech Alike
At Drover we have very clever people who are great at spotting problems and applying tech to solve those problems. To earn my place amongst their ranks I have been trying to explain a piece of tech that is very popular at the moment and have enjoyed some success.
Machine Learning is a great buzz term, it excites our imagination because of its inherent contradiction. Machinery is cold and metallic, whilst learning is something rosy cheeked school children do. Forget about the words themselves and just think of it like this.
Machine learning is about using the information we have and giving something our best guess.
The best way to learn what something is, is to do it yourself.
…

Now, we are going to look at the same kind of info about both ducks and dogs:
The Colour
The Noise the Animal is Making
The Number of Legs it Has
It’s Weight Between 1–100
If it Can Fly
What it’s Coat is like
Ducks

Most are White
They make quacking noises
One unfortunate duck has lost his leg
They are quite light
They can fly (shock)
They have feathers
Dogs

They are mostly Brown
They make a variety of noises
They are quadrupeds
They might be heavy, they might be light
They can’t fly (shock)
They have fur
Easy, Right? Try These then…
Example 1

Dog or Duck? Scroll down when you’ve made up your mind.
( Medium won’t let me have blank space so there are trailing periods)
. <- ignore these
.
.
.
.
.
.
.

This is unfair isn’t it. Higher weights are only associated with dogs and we did see a white duck. Boo.
More of our ducks from earlier were white….
But, we had so many missing values for this guy that really we had very little to go on.
.
.
.
.
Example 2
Try this one. Again, scroll when you’ve made up your mind.

.
.
.
.
.
.

That’s more like it. It can’t fly, but it’s brown, has two legs and is quacking.
The higher weight and lack of flight lean us toward dog, but the quacking is probably conclusive…
The poor fellow simply has his wings clipped which is sad but understandable.
.
.
.
.
.
Example 3
Same drill. Scroll when you’re ready.

.
.
.
.
.
.

This is really unfair!
A small dog that has lost two legs and has wheels that are making a squeaking noise.
This is difficult because from our information earlier it’s ducks that weigh little and have two legs.
The squeaking noise tells us nothing as we haven’t seen it in our training data.
.
.
.
.
.
Example 4
Now try this.

.
.
.
.
.

About time!
He’s a perfect example of a dog in terms of his feature space. Weighs a lot, can’t fly, fur coat.
The white colour might lean us back toward duck but those other factors combined form a very convincing ontology of a dog.
.
.
.
.
.
Example 5
Now try this.

.
.
.
.
.
.

Hard to be sure what to say about this guy. Although apparently he can fly.
White colour leans us towards ducks.
But he was heavier than our ducks generally have been. But also also he could fly! Which of these are more important when guessing duck or dog?
.
.
.
.
Example 6
Last one.

.
.
.
.
.
.

And there we have it. One final screw you to anyone trying to make a sensible guess out there.
In this final instance:
We were completely wrong & were looking at the entire problem in the wrong way.
.
.
That’s unfortunate. How do we fix this?
We don’t.
We accept that sometimes, often in fact we are wrong.
But better to be right most of the time than not try at all. We are trying to make something that is useful rather than perfect.
So what does any of this strange game mean
One of the problems we have at Drover is people leaving us. It happens all the time and it never gets any easier. If we can spot people before this happens and turn on the charm we may be saved some heartache.
Imagine instead of Ducks and Dogs we have Churners and Keepers.
Imagine instead of fur and flight we have time active and website activity.
We can begin to make guesses about who might be a churner and a well placed phone call could preserve a happy relationship and *ahem* keep us in business.

