Mastering AI/ML from First Principles - Part 2

• Updated on: • 6 min read • AI-ML

Numbers, Variables, and Why ML Needs Math

Series: Mastering AI/ML from First Principles
Phase: Math Foundations
Blog #: 02
Topic: Algebra for Machine Learning → Numbers, Variables, and Why ML Needs Math

1. Why this matters

When I started seriously exploring machine learning, I kept hearing the same advice:

“You need to know math.”

That statement always felt incomplete.

It didn’t tell me:

  • what math actually matters
  • how deep I need to go
  • or how it connects to real models

And honestly, most explanations either go too abstract or too superficial. So instead of taking that advice at face value, I decided to do something different.

I started looking at actual machine learning models—not implementations, but the equations behind them.

And what I realized surprised me:

At the core, machine learning is not complicated math. It is simple math used in a structured way.

Before gradients, before optimization, before neural networks—there are just:

  • numbers
  • variables
  • relationships between them

If I don’t understand these deeply, everything else becomes memorization.

This blog is where I slow down and build that foundation properly.

2. Intuition — What is really happening in ML?

Let me strip machine learning down to its simplest form.

Every ML problem looks like this:

“Given some inputs, predict an output.”

Examples:

  • Given house features → predict price
  • Given user behavior → predict churn
  • Given past sales → predict future demand

That means:

  • Inputs → numbers
  • Output → number
  • Model → relationship between those numbers

That’s it (in a simplified way).

Before thinking about models, I need to be comfortable with:

  • what numbers represent
  • how variables store those numbers
  • how relationships are expressed

Because machine learning is really:

turning the real world into numbers and learning relationships between them

3. Core explanation

3.1 Numbers — everything becomes numeric

This was one of the first mindset shifts I had to make.

Computers don’t understand:

  • “house”
  • “experience”
  • “good performance”
  • “high demand”

They understand numbers. So everything gets converted.

Real-world concept

Numeric representation

House size

1500 (sq ft)

Experience

5 (years)

Temperature

22 (°C)

Yes/No

1 / 0

Even things that don’t feel numeric—like text or images—are eventually represented as numbers. That means:

ML is not about objects. It is about numbers representing those objects.

That realization alone made things much clearer for me.

3.2 Variables — giving meaning to numbers

A number by itself is just a value.

A variable gives it meaning.

Instead of saying:

  • 1500
  • 300000

I say:

  • x = 1500 → house size
  • y = 300000 → house price

Now I can reason about it.

Variables allow me to:

  • label data
  • manipulate relationships
  • build models

This is where math stops being abstract and starts becoming useful.

3.3 Inputs and outputs — the ML structure

Every ML problem follows the same structure:

  • Input variables (features) → what I know
  • Output variable (target) → what I want to predict

Example:

  • x = house size
  • y = house price

Or:

  • x = study hours
  • y = exam score

This simple structure shows up everywhere.

3.4 Relationships — the heart of everything

Now comes the most important idea.
Instead of memorizing outputs, I define a relationship.
This equation looks simple, but it contains everything:

  • x → input
  • y → output
  • m → how strongly x affects y
  • b → baseline

When I first saw this in ML, I thought:

“This is just school math.”

Now I see it differently:

This is a machine learning model.

3.5 What changes in machine learning?

In traditional programming:

  • I define rules explicitly
  • Example: if x > 10 → do something

In machine learning:

  • I give examples
  • The system learns the relationship

So instead of choosing m and b, the model learns them from data.

But the structure itself?

Still algebra.

4. Real-world analogy

I started thinking of this like a business problem.

Let’s say I run a small coffee shop.

I want to estimate daily revenue.

Inputs:

  • number of customers
  • average spend

Output:

  • total revenue

A simple relationship could be:

Revenue = Customers × AvgSpend

That’s just algebra.

Machine learning takes this idea and makes it:

  • data-driven
  • flexible
  • scalable

But the core idea doesn’t change.

5. Worked examples

Example 1 — Study hours

Let:
y = 5x + 40
x = study hours

y = exam score
If x = 6:
y = 5(6) + 40 = 70

Interpretation:

  • Each hour adds 5 points
  • Even without studying → baseline is 40

Example 2 — Salary estimation

$$
\text{Salary} = 5000x + 40000
$$

  • x = years of experience

If x = 10:

This is exactly how early ML models behave.

Example 3 — Multiple inputs (preview)

Later, models become:

$$
y = w_1 x_1 + w_2 x_2 + b
$$

Where:

  • size
  • location
  • rooms

All contribute to price.

This is where ML becomes powerful—but it still starts here.

6. Math (only what I actually need right now)

Linear relationship

y = mx + b

  • m → rate of change
  • b → baseline

This equation shows up everywhere in ML.

What this leads to later

This one idea evolves into:

  • linear regression
  • neural networks
  • gradient descent

That’s why this matters more than it looks.

7. Python implementation

# Blog 02: Numbers and Variables

import numpy as np
import matplotlib.pyplot as plt

def predict_score(x):
    return 5 * x + 40

x = np.array([0, 2, 4, 6, 8, 10])
y = predict_score(x)

print("Study hours:", x)
print("Predicted scores:", y)

plt.figure(figsize=(8,5))
plt.scatter(x, y)
plt.plot(x, y)
plt.xlabel("Study Hours")
plt.ylabel("Score")
plt.title("y = 5x + 40")
plt.grid(True)
plt.show()

What this really means

  • x → input
  • y → output
  • function → relationship

This is not ML yet.

But it is the exact mental model ML builds on.

8. Key takeaways

  • Everything in ML becomes numbers
  • Variables give meaning to those numbers
  • Models are relationships between variables
  • Algebra is the foundation—not an optional skill
  • If I understand this well, ML becomes much easier to reason about

9. What confused me

I initially thought:

“Math comes later, after I learn ML.”

That was completely wrong.

Math is not something I “apply” after ML.

Math is the structure of ML.

The moment I stopped seeing equations as symbols and started seeing them as relationships, things clicked.

10. References

  • ISLR — Introduction to Statistical Learning
  • ESL — Elements of Statistical Learning
  • Hands-On Machine Learning — Aurélien Géron
  • NumPy Documentation
  • Matplotlib Documentation

 Next step

Next blog:

Functions — The Language of Machine Learning

That’s where:

  • models = functions
  • input → output becomes formal
  • graphs start mattering

And things start accelerating.