If you want to learn data science and don’t have a background in math, start here.
One of the most common questions I receive from data science aspirants is “how much math do I need to know for machine learning?”
When I first started teaching myself data science, I didn’t know exactly how much math I needed to learn in order to become fully qualified for an entry-level data science position.
I spent a lot of time studying undergraduate-level calculus, linear algebra, and statistics.
After I landed my first data science job, I realized that the amount of math required when working in the industry was far lesser than I’d expected.
Most companies tend to build machine learning models to solve very similar business problems. Due to this, you generally don’t need to re-invent the wheel and build algorithms from scratch. You can often use pre-existing solutions to fit your current use-case.
However, it is always better to have at least an intuitive understanding of the working behind these algorithms.
For example, gradient descent is used to find an optimal slope value in linear regression models. This is an application of calculus, and it is always beneficial to understand how the line of best fit is calculated.
Similarly, linear algebra has applications in data preparation for modelling, and is used widely in implementing dimensionality reduction solutions. If you want to expand your knowledge of deep learning, you will need to learn matrix manipulation techniques.
Finally, statistics is the most important branch of mathematics that you need to learn in order to become a data scientist. When working in the field, you will need to analyze trends in data, frame hypotheses, and transform observations into meaningful insights. These are all applications of statistics, and is a must-learn if you want to land a data science job.
When I learnt the concepts above, I initially spent a lot of time performing calculations by hand. I learnt to differentiate, integrate, and solve linear equations. However, this isn’t a pre-requisite to learning data science.
We have computer programs that will perform these calculations for us, and what we really need is to get a high-level, intuitive understanding of these mathematical concepts.
Again, the advice above only applies if you’re trying to learn data science to get a job in the industry. If you’d like to become a machine learning researcher or go into academia, then the amount of math you need to learn will significantly increase, as you will be working to build new solutions from scratch.
In this article, I will provide you with 6 resources to learn math for machine learning. I will include free online material (YouTube videos, online courses, textbooks) that you can use to gain enough proficiency at math to become a data scientist.
1. 3Blue1Brown’s Linear Algebra Series
3Blue1Brown is a popular YouTube channel that takes a visual approach to break down highly complex math concepts.
Their linear algebra series will take you through the core linear algebra concepts, such as vectors, linear combinations, linear transformations, matrix multiplication, eigenvalues, and eigenvectors.
They don’t go too deep into mathematical calculations. Instead, the focus of this series is to provide you with an intuitive understanding of linear algebra. Instead of simply memorizing formulas, you will be given an explanation as to why they work and how you can derive them yourself.
2. 3Blue1Brown’s Calculus Series
This is the second resource on this list created by 3Blue1Brown. Their calculus series is also intuitive and easy to understand.
They will walk you through concepts like derivatives, the chain rule, and implicit differentiation — all of which have a direct application in implementing the gradient descent algorithm.
3. 3Blue1Brown’s Deep Learning Series
While their calculus series is more general and catered towards anyone who would like to enhance their knowledge in math, 3Blue1Brown’s deep learning series is created specifically for students interested in artificial intelligence.
The deep learning series will take you through the inner workings of a neural network, and how they learn.
You will be provided with an in-depth explanation of the backpropagation algorithm and how it works, along with the calculus concepts behind it.
4. An Introduction to Statistical Learning
This is one of the best introductory statistics textbook that you can read as a machine learning enthusiast. It is available online and can be downloaded for free.
If you have implemented algorithms like linear regression, logistic regression, and decision trees in the past but don’t understand the working behind these models or when to use them, this book is a great place to start.
You will gain an intuitive understanding of how linear models, tree-based algorithms, and unsupervised techniques work. Explanations are provided as to how you can combat overfitting when creating different statistical models.
Every chapter in this book also includes a lab exercise in R that you can code along to, so you don’t just gain a theoretical understanding of the subject, but are also able to put the concepts learnt in practice.
5. Statistics 110 by Harvard University
The final resource on this list is a YouTube series called statistics 110, that has been made available to the public by Harvard University.
This is one of the best series of lectures I’ve found online for statistics and probability. It covers almost all the statistical concepts you’d generally see in an undergraduate level statistics class — probability axioms, types of distributions, the Monty Hall problem, covariance, correlation, Chi-Square tests, T-tests, Markov chains, and more.
The lectures start you off at an introductory level, and you can follow along with little to no difficulty even if you don’t have a background in statistics and probability.
The resources above are a great way for you to dip your toes into the world of math for machine learning. They provide you with an intuitive understanding of mathematical concepts, and this will enhance your knowledge of the working behind the models you build every day.
If you don’t have time to work through all the resources in this list, then I suggest at least reading An Introduction to Statistical Learning, as the concepts explained in this textbook have a direct application to your daily workflow as a data scientist.
Thanks for reading! This article was originally published here.