Traditional programming

In traditional programming, you create explicit rules.

• Input data + your rules → output

  flowchart LR
  I[Input] --> R[Rules (code)] --> O[Output]

Example: a tax calculator.

• You have a written policy.
• You implement the policy in code.

Machine learning

In ML, the “rules” are learned from data.

• Input data + output labels → learning algorithm → model
• Later: input data + model → predicted output

  flowchart LR
  subgraph Training
    D[Training Data (X)] --> A[Learning Algorithm]
    L[Labels (y)] --> A
    A --> M[Model]
  end

  subgraph Inference
    N[New Input (X_new)] --> M --> P[Prediction]
  end

What changes in your workflow?

Traditional programming focus:

• getting the logic right
• writing tests for rule correctness

ML engineering focus:

• getting training data right
• choosing features
• preventing leakage and overfitting
• measuring with the right metrics

A simple example

Traditional approach (rule-based spam)

def is_spam(email_text: str) -> bool:
    keywords = ["free", "winner", "click here"]
    text = email_text.lower()
    return any(k in text for k in keywords)

def is_spam(email_text: str) -> bool:
    keywords = ["free", "winner", "click here"]
    text = email_text.lower()
    return any(k in text for k in keywords)

ML approach

You collect labeled examples:

• email text → spam / not spam

Then train a model on those examples.

Key takeaway

• Traditional programming is rules-first.
• ML is data-first.

If you don’t have good data (or enough), ML usually disappoints.