How To Learn Machine Learning - the Right Way

Machine Learning (ML) has become really famous over this decade. It has been integrated with every known field out there, be it finance, crypto, trading, healthcare, analytics, robotics, cryptography, defence, or smart TVs at our home, and many other products that we use in our day to day lives. Machine Learning has changed our lives and it will be used even more in the future.

When I first started, I was also confused reading and watching lots of tutorials and youtube videos to learn ML in a structured way, but at the end I started doing things on my own, mostly by doing hands-on and projects in ML. But I also got stuck frequently and had to take breaks to understand the fundamentals and theory. This article covers everything that I feel I should have known before starting ML.

ML is a very vast field and one may feel overwhelmed at the start. As a beginner in ML, you must be confused where to start from – should I start making ML predictive models in Python right from the start or learn basic mathematics like statistics and probability, and if so then till what extent should I learn these topics to call myself one level higher than an ML beginner?

All these overwhelming questions will be answered ahead. Let us see what we are going to read further.

Table Of Content

1. What is Machine Learning
2. Step wise guide
1. Basic Mathematics and Programming
2. How much ML is required?
3. Projects - get a hands on experience
4. Join ML Communities
5. Explore ML Further

Beginner's Guide to Machine Learning

What Is Machine Learning

As the name suggests, it is the formula or a way through which a machine starts to learn, identifying patterns to guess over unknowns of the same kind.

In other words, it is the learning process of machines.

I would recommend following the 80 by 20 rule. It says whenever you learn something new, give 20% of the time to read and understand the theory aspect and 80% of the time to do practical implementation. When you make mistakes and debug, you learn hands on which is more effective, interesting and you will remember the small points without even trying to learn.

Step 1: Learn the Mathematics for ML

This step is important because everything in ML is based on calculations, and most of the time you will see vectors and matrices. Below is the structured way to learn the relevant topics and the top sources to learn from.

In Linear Algebra, you should cover vectors, matrices and their operations like dot product, cross product and matrix multiplication as well as all the properties of matrices.

Tensors will be very useful when you will learn deep learning algorithms. Tensors are just stacked matrices, just as matrices are stacked vectors. Tensors have 3 dimensions, matrices have 2 dimensions and vectors have 1 dimension.

EigenValues and EigenVectors are also important topics that are used in Principle Component Analysis (PCA) and Dimensionality Reduction techniques.

Matrix Decomposition is also a very useful concept that is useful for data compression techniques such as Singular Value Decomposition (SVD).

Overall, I think this is the minimum you should know in Linear Algebra before marching ahead.

1. Vectors and its operations.
2. Matrices and its operations.
3. Tensors.
4. Eigenvalues and EigenVectors.
5. Matrix Decomposition.

The above topics are most relevant and will be used frequently while you do hands-on learning on ML projects.

It’s because data in ML is often represented as vectors and matrices (especially images) and in the real world or 3D scenario, tensors will be used.

Algorithms like linear regression and neural networks rely heavily on matrix operations for calculations and you will use eigenvalues and eigenvectors to perform PCA on large datasets.

That’s all linear algebra to get you started, but it’s recommended to get a solid understanding of every topic and learn new topics consistently.

At this point, you can start hands-on with Python’s Numpy library. Be comfortable with functions in numpy such as np.dot() for dot product, np.matmul() for matrix multiplication, np.linalg.eig() for eigenvalues and eigenvectors, etcetera.

You can start learning from Khan Academy lessons on Linear Algebra or from any linear algebra book.

Linear Algebra and Its Application by Gilbert Strang.

In Statistics and Probability, you should cover descriptive statistics, probability basics, probability distributions, bayes theorem, hypothesis testing, expectation and variance and correlation and covariance.

In depth grasp of these topics will make you robust in identifying the gaps in datasets and analyzing how efficient an ML algorithm or model is.

Statistics will let you know about how balanced the dataset is and analyse the data for outliers and normalizing features. Probability is the integral part of ML algorithms that uses predictive ability such as logistic regression for classification.

Distributions are used to initialize model parameters such as Gaussian Mixture model uses after analyzing the dataset.

You can start learning from Khan Academy lessons on Probability and Statistics to clear your basics. You can also learn in a traditional way through books to get a deeper understanding of the subject.

Probability book

Probability and Statistics book

Now you are ready to understand and handle datasets. Use pandas or polars in python script to analyze and perform operations on a dataset.

At this stage, I would recommend going on Kaggle to hone your skills on a variety of datasets available on the platform.

Below are some beginner friendly datasets, you can practice and analyze these datasets.

1. College Student Placement Factors Dataset, access dataset.
2. FlavorSense Predict Taste by Life and Climate, access dataset.
3. Data Science, AI and ML Job Salaries in 2025, access dataset.

You can try a classification algorithm on these dataset or just practice visualization skills to derive meaningful insights from the data.

Calculus is also very important because it tells us how to optimize an ML model. When you will go deeper in Machine Learning and then Deep Learning, you will see that many algorithms use optimization by reducing the loss functions. It became possible only because of calculus.

In calculus, you must at least learn functions, derivatives, partial derivatives, gradients, gradient descent and basic understanding about how integrations work which will be useful later on while learning probability density function.

1. Functions
• Single and Multi variable functions.
2. Derivaties
• focus on chain rules
3. Partial Derivaties
• core concepts in neural networks.
4. Gradient and Gradient Descent.
5. Integration.

You can learn these topics from Khan Academy lessons on Calculus 1 and 2, especially for derivatives and integrals.

Geeks for geeks is really helpful once you have grasped the theoretical concepts, geeks for geeks calculus in python.

Calculus for Machine Learning and Data Science by deeplearning.ai is also a great course on Coursera.

If you want to further deepen your understanding in calculus, you can refer to this book.

Deep Learning by Ian Goodfellow.

Machine Learning book by Christopher M. Bishop.

Now you should try to implement these concepts in python such as using gradient descent to minimize a quadratic function in code or if you are confident in yourself, you can create your own algorithm to minimize the error.

Learn a Programming Language

Python is the best choice for doing ML programming because it has lots of libraries prebuilt for Machine Learning tasks and has a very big community of ML users.

Python is a very user-friendly programming language. You will learn this language quite easily, it’s just that learning various kinds of python libraries, modules and their functions will take you quite some time. But you don’t need to worry about it, because even after a lot of practice and experience, I still need to google and refer to documentation whenever I need it.

When you study mathematics for ML, start doing the relevant hands-on code in python parallely. You are free to explore other things but it is recommended from my side to cover the above mentioned topics before too far.

You can learn Python from Coursera, YouTube, freeCodeCamp, or Python books.

If you know some basics of python programming, you can go with Data Analysis with Python by IBM on Coursera.

In case you are a complete beginner in programming, start with Get Started with Python by Google on Coursera.

There are many free websites to learn programming languages like W3Schools, geeks for geeks, codeacademy, leetcode, etcetera.

Step 2: Machine Learning Concepts

This includes the basics of Machine Learning and its types. Machine Learning is divided in 3 categories based of how they learn from data and solve the problem, these are:

1. Supervised Learning
• Regression
• Classification
2. Unsupervised Learning
3. Semi-supervised Learning.

After learning their underlying theory, go to different kinds of algorithms that come under them, such as linear regression under regression type supervised learning, logistic regression under classification type supervised learning, k-means clustering which comes under unsupervised learning, etcetera. There are many more, you can explore them from basic to advanced.

Every type has traditional and novel algorithms, novels are those that use neural networks while traditional doesn’t. For example in unsupervised learning k-means clustering is a traditional algorithm, while autoencoders is a novel clustering algorithm.

To learn the basics of ML in the shortest time, in depth and top quality, I would say to follow the courses of Dr. Andrew Ng, the founder of Deeplearning.ai. His courses are of top quality, precise and very interesting, you will definitely learn a lot from him.

You are free to read various Machine Learning books, to strengthen your understanding in the topics even deeper. This book from Mr. Chip Huyen is very good for beginners who want to go deep afterwards.

Machine Learning book by Mr. Chip Huyen.

Step 3: Get Hands-On with Projects

Getting hands-on with projects is really a great experience. You will get to learn a lot more while creating your own project from scratch. The best thing is that you will get errors and it will teach you more than what theory can teach you.

You can do sample projects first available on kaggle like Titanic Survival Prediction, or with the default Iris dataset available in the iris module.

It’s your choice where to start but start with projects that don't seem too complex at first. There are many more beginner friendly projects on kaggle.

I would personally recommend you to participate in the monthly Kaggle Playground series. It is a great place to hone your skills. New competitions are launched every month in the Kaggle Playground series. Many seasoned professionals and data scientists also participate in this from all over the world. Every competition has a dedicated page for discussion and you can start your own discussion as well. It’s really a great community to learn Machine Learning and Data Science.

Step 4: Join an ML Community

Kaggle is the biggest community of Machine Learning and Data Science enthusiasts. I have already mentioned you guys to practice on Kaggle. Along with practicing on kaggle, you can follow some really great subreddits on Machine Learning, only if you use reddit.

One such ML subreddit is reddit.com/r/learnmachinelearning.

Use discord to stay updated and discuss with ML enthusiasts for upcoming or ongoing ML competitions. Kaggle is also on discord, you can join them.

Let’s not try to make this too complicated, it’s already a lot of information for a beginner. You can try to follow this for a few weeks, you will definitely see the improvements. Just be consistent and honest with yourself during the process.

Document your journey, it will be really motivating during the process. If you are comfortable, you are welcome to share your journey here on PhyCosmos or me personally through email.

My email is: shivamgravity@gmail.com.

Ask questions, share your projects, and don’t be afraid to make mistakes. Everyone’s learning!

Step 5: Keep Learning and Experimenting

That’s all you need to start your Machine Learning Journey, just take one step at a time. Don’t try to go too fast, be at your own pace. Learning basics is important so that you don’t implement wrong things during hands-on.

You can build your own projects, with your own idea from scratch. You can participate in various competitions and hackathons happening worldwide. Just google them and you will find a lot of competitions going on.

Just a reminder to always fill your plate with what you can eat!

Otherwise, you will get overwhelmed and won’t meet deadlines for any of the projects. It’s really bad practice, and I have done this a lot of times and felt regret not to focus only on a few that I can handle. If you are a beginner, this is the best advice I can give you.

It doesn’t mean to be afraid to participate, or thinking like you will miss something big. I want to say that you should be aware of your limits and increase them gradually over time.

Still, if you think you can manage lots of tasks. You can take on the challenges. If you have successfully done this as a beginner, please share with us your insights about managing lots of projects at a time. I really want to know about your journey and methods you followed.

Once you got comfortable with the process and feel confident enough on the basic, move to other domains like:

1. Neural Networks and Deep Learning.
2. Natural Language Processing (NLP).
3. Reinforcement Learning (a very interesting domain, for sure).

I have included all the advice, tips and tricks in this article to get you onboard smoother on ML Journey, than I did back as a beginner.

I hope you have gained something different and valuable from here. I have written this article based on what I have faced while starting Machine Learning as a beginner 2 years ago.

If you have any query, you can reach out to me via email.

Thanks a lot for supporting me!

Note: Some links in this post are Amazon affiliate links. This means I may earn a small commission if you make a purchase, at no additional cost to you.