What's the right tool for the job?
Machine Learning is a tool Product Managers can leverage
I’m endlessly eager to learn new things and apply them to real problems. I think this is a natural part of my character. As a young child, I remember wanting to dig to the center of the earth with my brother. He and I had a plastic sand bucket with a matching bright yellow hand rake.
One day as we began to scratch the earth with our plastic tool, it quickly became clear that it was going to take more than our determination to make it to the molten rock under the earth’s crust.
Knowing we weren’t supposed to touch my dad’s tools we smuggled screwdrivers, hammers, wrenches, sockets, pliers and spark plug gappers to help us with the job.
I was feeling pretty good about the process as we used the hammer and screwdriver to chisel the dirt. While the spark plug gapper proved to be less than help in measuring the progress we were making. Given our time, budget, and resource constraints, we had found the right tools for the job.
So with that in mind, let me suggest to you that…
ML algorithms and models are tools
Broadly, machine learning (ML) is a branch of AI and computer science that focuses on using data and algorithms to make predictions or learn without explicit programming.
The field of ML relies on ML algorithms to build ML models.
“Machine learning algorithms improve performance over time as they are trained—exposed to more data. Machine learning models are the output, or what the program learns from running an algorithm on training data. The more data used, the better the model will get.” (source)
Pro tip: When you hear “machine learning” or “ML” ask yourself is it in reference to the field, algorithm or model?
To re-cap, Machine learning algorithms and the models they support are a tool Product Managers can use to build solutions for the users and systems they serve. But…
How do you know what type of tool you have?
To know what type of tool you have you must dig a level deeper… The field of ML relies on ML algorithms to build ML models. The ML algorithms are then categorized as supervised or unsupervised.
In a supervised model, the data is “labeled”. For those of us who’ve been PMs for some time we might liken this to tagging a data set with metadata.
For example, if the data point you have is the pixels of an image, and it is labeled as “cat”, “dog” or “cow” through some type of process - off-shore resources, automation, crowd-sourced data creation - it doesn’t matter really. Then your data set is “supervised”.
In an unsupervised model, the data is not labeled. It looks at historical values to predict future values.
Let’s look at an eCommerce recommendation system example
Let’s say we want to build a recommendation system and this is the data set. It can be used to predict whether a user is likely to purchase a specific product based on their demographic information and past purchase behavior.
This unsupervised dataset can be used for the same purpose—recommendations—but its approach could differ. For example, a ML clustering algorithm could group users with similar shopping behaviors and demographics. It would do so by identifying patterns in the data without needing labeled data such as purchase history.
Why is it important to know if a model is supervised or unsupervised? It tells you what type of “tool” the ML algorithm is. Some tools will be better suited than others for specific jobs.
Next week I’ll share with you a simple set of questions you can use to categorize your data sets. And soon after I will share practical advice on how to evaluate which model is best suited for your application.
It never hurts to try
In case you were wondering, my mother caught us red-handed with my father’s tools, so we never reached the center of the Earth. We did, however, dig a hole big enough to be the envy of other kids in the neighborhood ;)
Share Your Experience
Did you ever use your parent’s tools? What problem did you solve with it? What did you learn from the experience?
As a Product Manager, when it comes to Machine Learning what do you want to learn more about? What problems would you like to solve using ML algorithms or models?
What was the most helpful thing you learned from this article?
I’m interested in your thoughts. Share in the comments below.