Although mathematical optimization can be a force multiplier for machine learning and data science, there is a general lack of awareness and quite a few misconceptions around what it is and what it can do.
In my role as a data science strategist at Gurobi, I try to break down some of the barriers that may be keeping data scientists and the broader AI community away from optimization.
I joined Jon Krohn for an episode of his Super Data Science podcast to discuss the basics, including how data scientists can leverage this powerful tool.
Below are some highlights from our conversation.
As I shared with Jon, we can think of data science, machine learning, and mathematical optimization as different tools in a toolbox. Sometimes, you need a hammer (which might be data science), while other problems require a screwdriver (which might be mathematical optimization).
It all depends on what you’re trying to accomplish. It’s all about selecting the right tool for the right job.
You may even get pretty far by building out a custom methodology for decision making, but from a business perspective, if you’re not using the right tools for the job, you could be leaving a lot of money or efficiency gains on the table. For example, cutting fuel costs by just 1% more could potentially mean millions of dollars in savings.
That’s where mathematical optimization comes in. Whereas machine learning is great at predicting, optimization tells you what actions are needed to get you the best results. It can help you navigate your most complicated decision problems and prescribe next best steps based on your objectives.
Mathematical optimization guarantees two things: First, it guarantees feasibility, which means that once you set up all of your constraints and decision variables, you can be sure that your constraints will be satisfied.
Second, you can count on global optimality, which is unique to mathematical optimization. While you could use a heuristic approach, you may get stuck in a locally optimal solution. That’s why optimization is essential when you’re looking for the best solution, not just a best guess.
So what does all this look like in practice?
First, you must translate your business problem into a mathematical model (typically a linear programming model or mixed integer programming model). That model must then be translated into code. A solver, like Gurobi, then uses advanced algorithms to find the best possible solution based on your defined constraints and objectives.
While open-source solutions can work for more simple algorithms, a specialized software is often needed to handle more complex, real-world problems.
Throughout our conversation, Jon and I also discussed one of the biggest inhibitors to mathematical optimization’s growth as a field, which is the notion that you need a PhD in operations research in order to utilize it.
At Gurobi, we constantly seek new ways to lower that barrier to entry. While having a mathematical background certainly helps, we make an effort to introduce mathematical concepts step by step and help people understand what exactly is going on.
For example, we provide access to a library of Jupyter notebooks, including some that are aimed specifically at the data science crowd. These notebooks feature more simplified math, along with explanations of notations, which makes it easier to read through the problem.
It’s part of Gurobi’s mission to not only innovate but also to educate, ensuring that everyone has access to the knowledge they need to harness the power of mathematical optimization.
That’s why you’ll find a wealth of free resources on our website that are designed to help data scientists and those who may not have a strong mathematical background get started.
For an even closer look at optimization, you can listen to my full conversation with Jon on the Super Data Science podcast, where we explore in detail:
Jon and I will continue our conversation in August, when we explore Gurobi’s interactive game that can help anyone learn optimization, plus a free Udemy course and more on our Jupyter notebooks—stay tuned!
Data Science Strategist
Data Science Strategist
Mr. Yurchisin has over ten years’ experience applying operations research, machine learning, statistics, and data visualization to improve decision making. Before joining Gurobi, Jerry (who also goes by Jerome) was a Senior Consultant at OnLocation, Inc. where he customized several linear programming models within the National Energy Modeling System (NEMS) to analyze implementing specific energy policies and utilizing new technologies. Prior to OnLocation, Jerry was an Operations Research Analyst & Data Scientist at Booz Allen Hamilton for over seven years. There he formulated scheduling and staffing integer programming models for the US Coast Guard, as well as led a project to quantify the maritime risks of offshore energy installations with the Research & Development Center. Further, Jerry was the technical lead on several Coast Guard studies including Living Marine Resources and Maritime Domain Awareness, providing statistical analysis and building supervised and unsupervised machine learning models. He also performed statistical analyses, machine learning modeling, and data visualization for cyberspace directorates at DoD and DHS. Jerry has several years of experience teaching a wide variety of college-level mathematics and statistics courses and has a passion for education. He also enjoys golfing, biking, and writing about sports from an analytics point of view. He lives in Alexandria, Virginia with his wife, son, and two dogs. Jerry holds B.S., Ed. and M.S., Mathematics degrees from Ohio University and an M.S. in Operations Research and Statistics from The University of North Carolina at Chapel Hill.
Mr. Yurchisin has over ten years’ experience applying operations research, machine learning, statistics, and data visualization to improve decision making. Before joining Gurobi, Jerry (who also goes by Jerome) was a Senior Consultant at OnLocation, Inc. where he customized several linear programming models within the National Energy Modeling System (NEMS) to analyze implementing specific energy policies and utilizing new technologies. Prior to OnLocation, Jerry was an Operations Research Analyst & Data Scientist at Booz Allen Hamilton for over seven years. There he formulated scheduling and staffing integer programming models for the US Coast Guard, as well as led a project to quantify the maritime risks of offshore energy installations with the Research & Development Center. Further, Jerry was the technical lead on several Coast Guard studies including Living Marine Resources and Maritime Domain Awareness, providing statistical analysis and building supervised and unsupervised machine learning models. He also performed statistical analyses, machine learning modeling, and data visualization for cyberspace directorates at DoD and DHS. Jerry has several years of experience teaching a wide variety of college-level mathematics and statistics courses and has a passion for education. He also enjoys golfing, biking, and writing about sports from an analytics point of view. He lives in Alexandria, Virginia with his wife, son, and two dogs. Jerry holds B.S., Ed. and M.S., Mathematics degrees from Ohio University and an M.S. in Operations Research and Statistics from The University of North Carolina at Chapel Hill.
GUROBI NEWSLETTER
Latest news and releases
Choose the evaluation license that fits you best, and start working with our Expert Team for technical guidance and support.
Request free trial hours, so you can see how quickly and easily a model can be solved on the cloud.