Learning machine learning is much like learning any new skill. It is dependent on factors such as your existing knowledge of machine learning topics, computer literacy, attitude, and the time you have to devote to learning it. With that in mind, here are some guidelines.
To learn machine learning, expect it to take 6 months if you dedicate 40 hours a week to just learning it. Expect to study 2 years at 10 hours a week or through practical experience weaved into your job to have a solid foundation with machine learning.
This can seem daunting, but remember, learning a skill comes in stages. For example, you’ll start understanding the nuances of the software and picking up some of the terms when you are in the initial stages of learning, and you’ll be able to begin communicating better with data colleagues even here. It’s like being able to greet others in a foreign language and ask, “Where is the bathroom?”
Machine Learning Definitions + Types (Simplified)
Let’s start with what machine learning is as that will help understand how long it will take to learn it and at what stages you’ll feel fluent.
Put simply, machine learning is applying computer algorithms so that they learn from historical data to make decisions based on the experiences in that data.
You take rows of data, plug it into the amazing tools that are at your disposal, and it turns it into a graphical output or number to tell you if it is a good model or not. If you know where to look, the fancy algorithmic models will tell you if it did a good job!
There are 3 types of machine learning, and we’ll talk about them in the order of easier to harder to learn. These are high level definitions. Understand that as we move from top to bottom, you move more from a data analyst skillset to more in the scope of data scientist.
Supervised learning is where you’ll start with machine learning. Supervised learning means that all of the columns have headers, and all of the rows have an outcome identified. An example is feeding a model with 1,000 customers who didn’t pay their medical bill and 1,000 customers who did pay their medical bill. You’re telling the model that this subset of the data had which behavior with no guesswork on what you’re feeding it. Supervised learning algorithms examples are linear regression, decision trees, and even neural networks.
Unsupervised learning is the next layer of machine learning. In unsupervised learning, you do not have to have the columns completely labeled or the outcome labeled for each row. The algorithm looks for patterns or anomalies to emerge without your intervention. Example algorithms for unsupervised are Gaussian Mixture or Self-Organizing Maps.
Lastly, there is reinforcement learning which is where the models continually take in feedback on their performance and morph accordingly. Were you good in school today? Here’s your green sticker. Did you predict poorly? Red sticker. This is definitely simplified, but I rarely see this type of data work in business organization as it is difficult to execute for those environments. I didn’t say never – I said rarely. This is what is powering self-driving cars, after all.
Is it Hard to Learn Machine Learning?
As you move down the types of machine learning, the underlying knowledge you need to acquire in the overarching computer science skillset increases.
Learning machine learning can be difficult. If you have a background in working with data, statistics, or even programming, you will find machine learning less intimidating.
That said, you don’t have to know EVERYTHING there is to know to succeed in learning aspects of machine learning to apply to the problems you need to solve. As I mentioned, I rarely come across reinforced learning in a business setting, and truth-be-told, it’s not often I have to dust off my unsupervised learning skills to help organizations solve problems. The sweet spot in leveling up data use has been in the supervised learning space.
Coincidentally, learning about supervised machine learning is where you’ll start, and you may have already seen the terms like linear regression or decision trees somewhere in your past. My 9th grade son is working with regressions now even though I don’t recall that knowledge in high school myself!
Can I Teach Myself Machine Learning?
You may have noticed the overload of available of online courses, books, boot camps, and university courses available to teach you machine learning. When I see ‘teach myself’, I’m assuming you mean that you want to teach yourself with the support of an self-paced online course or book.
You can teach yourself machine learning with the support of training materials such as online courses.
More often than not, I see courses on machine learning that are associated with learning software such as Python or R. You can certainly pick a language and go, but you can also learn the principles of machine learning without learning a language. It will be difficult to practice, but if you are looking to understand processes as a business leader looking to ask the right questions, you have options, too. Here is a machine learning course that I recommend when someone asks via Udemy.
Do I Need to Be Good at Math for Machine Learning?
To get started with machine learning, you do not need to be good at math. Honestly, the tools you have to apply data knowledge mean you just need to be able to survive the software. That said, you can’t operate without math in a silo. You need to understand what levers you have to pull that apply to your situation.
An example is that Power BI has developed a tool called Key Influencers that is using regression if an outcome based on a number and a decision tree if based on a word (yes/no). You do not have to be a machine learning whiz to use data tools, so it feels super approachable to put that kind of function in there. HOWEVER, when I used it recently, it wasn’t accounting for something called a rare event (when the outcome doesn’t happen often). Power BI doesn’t give me any look under the hood, so I had to figure out what was going on by using Python. So that function is available, but you have no idea what it is doing or assuming based on your data. What about null values? Is it throwing the record out? No clue.
What does this have to do with math? Well, the levers you have to pull in using an algorithm are heavily driven by math, so even if you need to Google search your way through a problem, it is helpful to understand components to make the right decisions in the context of the problem you are trying to solve with machine learning.
How Many Hours a Day Should I Devote to Learning Machine Learning?
How many do you have? I’m only kinda kidding. To learn machine learning, it is a best practice to devote no less than 30 min a day. You can certainly spend more time if you are able to, but the minimum is so that you keep the repetition up. Gaps in learning at least on a daily basis mean that you aren’t getting the repetition in which will increase the likelihood that you forget what you’ve already learned!
The best situation to get up to speed would be to use real-world problems as fast as possible. If you are in a business intelligence or other role where you can incorporate the principles or algorithms you’re learning, you’ll find the learning curve much lighter. Real world data is messy and likely more relevant to you – both make for a deeper understanding of machine learning!
What is the Easiest Machine Learning Language to Learn?
I do not agree with the R vs Python vs nothing else conversations. They both have pros and cons even from a machine learning standpoint. If you can write code in 1 system, you can take a quick training or search your way to the same functions in another system. Here are a few thoughts to help you decide if you’re unsure:
Python is a language that is used for more than just data science pursuits, so if you have a background in it from other experiences, you’ll appreciate the quicker learning to curve to using the software.
If you do not have a background in any language, I’d go for the one you would have the most support in. If your company is using Python, even if you’re not in a role to use it yourself, you are aspiring to be, and colleagues will be your first line of support. Well, second, behind your preferred search engine.
How Can I Learn Machine Learning Faster?
Learning any new skill takes time, patience, and practice. And lots of patience. Give yourself some grace, but if you need to cram machine learning knowledge, it can be done as well as a crash course of any foreign language.
You can learn machine learning faster by immersing yourself in machine learning trainings, industry publications, and practice by applying those skills.
Udemy style courses like this one can be taken as fast or slow as your schedule allows, and you can even speed up the video if you can still understand it. In addition, if speed is of the essence, I recommend also consuming either print books, audiobooks, or podcasts on machine learning topics in times when training is not feasible. Full immersion will help with the speed to learning.
Just don’t forget that repetition and applying the skills are still needed to fully adopt a new skill, much less, master it!