ML Was Hard Until I Learned These 5 Lessons! (You Need This)

Here’s How I Keep Up with Continuous Learning in the AI/ML Trend

Aug 26, 2024

Source: Photo by Annie Spratt on Unsplash

Machine learning is not everyone’s cup of tea. It’s intimidating. The heavy math, the complicated code, and the huge knowledge burden make it seem like an uphill battle.

But what if I told you there existed lessons that could make your journey smoother?

From the vantage point of having spent more than 4 years studying machine learning, here are five key insights I wish I had known from the start.

Let me share them with you, so hopefully you don’t have to struggle as long as I did.

Lesson #1: How to Understand Math Formulas

I remember when I was first learning machine learning: I used to stare at the complex formulas, waiting for some kind of magical “aha” moment.

It never did.

The thing is, I was totally wrong in the way I was going to study math; I was setting myself up for failure from square one.

The secret? Don’t think of math as abstract formulas. Think of it as a translation of human ideas.

Researchers or Scientists do not even think in pure mathematical language. They have ideas in natural language just like you and me, but the math is to give form and execution to such ideas.

This is how you approach when you see a formula:

Step back and try to understand the human idea behind it.
Interpret each part of the formula as a piece of that idea.
Remember that operations such as sums and products are really nothing more than loops; conditions represent something along the lines of if-else statements in code.

This can often be the way to see the scariest math ideas transform into really doable ones.

Lesson #2: How to Understand Derivations

Have you ever observed a professor who goes off on a “derivation spree,” writing step after step, leaving you in the ideological dark of confusion and overwhelm?

Surely I have.

But here’s a secret I know: Each step in a derivation is simply the application of some particular rule or definition.

To master derivations:

Create a list of mathematical rules and definitions.
Practice recognizing patterns where these rules apply.
Go through derivations doing “pattern matching” with your list.
With practice, you’ll start to memorize common patterns.

You must know, it’s not about an innate mathematical genius. It’s about how you build your toolkit and learn to apply it effectively.

Lesson #3: How to Code for Machine Learning

Learning to code for machine learning can be a rollercoaster.

At first, it’s following tutorials — incredibly you’re making some progress and starting to implement some cool stuff.

However, once you attempt to build something on your own, you hit a wall. All of a sudden you waste hours on a few lines of code and feel frustrated and incompetent.

Writing code that works perfectly, the first time is rare.

Hours of debugging are a normal and expected part of coding. Coding became much less intimidating, once I accepted this. I knew the struggle was part of the process, not a sign of failure.

Pro tip:
Use tools like GitHub Copilot or ChatGPT or Claude to generate and explain code. They can be amazing learning aids. I specifically, prefer Claude for coding.

Lesson #4: How to Tackle Large Codebases

It can be actually intimidating to reach the very large codebase for the first time.

Strategy:

What has the strategy been? This is the strategy I wish someone told me long ago:

Find the main entry points in the project; in most ML projects, these are train.py, eval.py .
Set a breakpoint at the start of these files.
Invoke a debugger and start stepping through the code.

It is more like being able to take a guided tour inside the code base. You will see data preprocessing, training loop, model architecture, and evaluation metrics. It’s really powerful to take on an arbitrarily complex project.

For learning a specific algorithm, look for small, educational implementations instead of large optimized ones.

Again, use debugging to walk through the main functions and understand the core ideas.

Lesson #5: How to Master Machine Learning

The final and, in fact, most important secret: With ML, it takes time to get good at it, and that’s okay.

Most people drop off in ML due to undue expectations. They feel they have to grasp everything very fast and master it within two weeks.

The harsh truth is, it’s a long journey.

I started my journey in AI 4 and a half years ago, and still, I learn something new every day.

Remember:

It’s quite normal to miss a lot of things at first.
Failed interviews and projects are the way.

Once you accept that it’s going to take time, you will be more relaxed and start enjoying the process. And that is how you achieve success in this journey of learning.