Getting Started: From Zero to Machine Learning

If you see real value in bringing machine learning to your organization and you’re not sure how to begin, Spinnaker is here to help. Here’s a primer on what you need to know to take your organization from zero to machine learning.

Machine Learning: Setting Your Expectations

When it comes to machine learning, think baby steps. Once equipped with the basics, automated machine learning tools can provide insights for your existing projects in a matter of minutes. While they’re fast, however, don’t expect these quick answers to solve your biggest problems overnight. So what can you expect? Machine learning will give you insights driven by data that is not easily discoverable with intuition alone.

Once you’re familiar with the power of machine learning and the speed of results in the hands of novices, you’ll understand the magnitude of what you can achieve with an experienced, well-trained analytics professional at the helm.

Choosing a Programming Language

Like all technological developments, machine learning is experiencing a rapid evolution, not just in the public availability of more and more powerful algorithms but also in the platforms available to run them.

One programming language, R, is likely a new-to-you language, but the software has actually been around for 25 years and is well-known among statisticians and data miners. R is open-source—thus it’s free—and most of the code you’ll ever need for basic analytics is already publicly available on the web. This makes R a great place to start.

Beyond R, there are a number of other tools and languages available, as identified by KDnuggets, a popular website and newsletter on artificial intelligence, data science, and machine learning. Get the latest in a KDnuggets article about its 19th annual Software Poll: “Python eats away at R: Top Software for Analytics, Data Science, Machine Learning in 2018: Trends and Analysis.” 

In addition, a number of companies, like rapidminer and DataRobot, are entering the machine learning space, offering products and platforms that seek to integrate data prep, model development, and model deployment.

What you choose is ultimately up to you but keep this in mind: There is some stickiness to the platform you start with, so make sure your organization is on board.

Understanding Off-the-Shelf Algorithms

With off-the-shelf machine learning algorithms, you can start solving problems about your existing datasets within minutes. Before you decide which is right for you, you need to understand the type of problem you’re trying to solve. At its most basic, there are two types of readily available algorithms:

1. Algorithms designed to predict an outcome

These algorithms help explain an outcome on a dataset for which you already have a history of results. They’re classified as “supervised learning” algorithms. Your objective is to use the available data to predict an target variable—a process frequently leveraged for predicting future outcomes and the likelihood of certain events.

One subset of supervised learning will be of interest to those already familiar with logistic regression models. In fact, much of what your statisticians have done in the past still needs to be done when you leverage a machine learning algorithm. The big difference in machine learning is that the measure of success is less about the significance of each variable you use, but instead is about the performance on datasets and business problems that were not used to build the original model.

2. Algorithms designed to group populations with no predefined expectations

These algorithms help improve classification and create clusters. Their purpose is primarily to help you understand your population and improve targeting. This machine learning practice is classified as “unsupervised learning” because there are no targeted outcomes. If you use cloud-based services to watch movies or listen to music, you can bet one of these algorithms is responsible for deciding what you’re enjoying.

Don’t Try This at Home (At Least Not Alone)

So now that you know more about getting started, my final piece of advice is: Please don’t go it alone. I’m not saying you shouldn’t dive in and start using machine learning, but there is great benefit to strengthening your team by bringing in experienced analysts. The skills to leverage machine learning are becoming more democratized by the day. While you don’t necessarily need to hire a data scientist, you should be looking for experience. Having someone in your corner who is already well-versed in best practices and has learned from their own mistakes will be extremely valuable to your organization.

One more thing: When adding to your team—whether with an FTE or a consultant—I recommend looking for someone who brings a specific attitude as well as experience and skill to the table. You need someone to guide you who genuinely embraces the opportunity to help and teach others. By finding the one “right” person, you and your organization will be well on your way to developing a valuable machine learning competency that will benefit your business now and in the future.

 

Want to learn more? Keep an eye on Spinnaker’s blog and follow Spinnaker on LinkedIn to receive updates in your LinkedIn feed. For more specific questions about machine learning and analytics, contact Spinnaker Principal Brett Ludden directly at brett.ludden@spinnakerconsultinggroup.com.

0 Shares