🤖 Machine Learning For Beginners

A hands-on, from-scratch journey through the foundations of machine learning — built one concept at a time, in plain NumPy and Matplotlib, with no black-box libraries hiding the math. This repo follows the path from a straight line through to classification, implementing each idea by hand to really understand what's going on under the hood.

The philosophy here: build it yourself first, then trust the library. Every model is coded from scratch before scikit-learn is ever introduced.

---

🗺️ The Learning Journey

The project grows in three stages, each building on the last:

Linear Regression  ──►  Multiple Linear Regression  ──►  Classification
   (one feature)            (many features)               (0 or 1 labels)

---

📈 1. Linear Regression — linearRegression/

Predicting a continuous number (like house price) from a single input.

File	What it teaches
machine_learning.py	The model f(x) = w·x + b, training data, predictions
cost_function.py	Measuring error with the squared-error cost
costfunction_visualisation.py	Seeing the cost as a curve we want to minimize
gradient_descent.py	Learning w and b automatically by descending the cost

The model — a straight line with ML naming:

f(x) = w·x + b

Symbol	Meaning
x	input feature
w	weight / slope
b	bias / intercept
f(x)	prediction

💡 Key insight: "training" just means using gradient descent to find the w and b that make the cost as small as possible.

---

📊 2. Multiple Linear Regression — multiple-linearRegression/

Same idea, but now each example has many features, so the math goes vectorized.

File	What it teaches
multiple_linearRegression.py	Extending to f(x) = w·x + b with vectors
vectorization.py	Using np.dot instead of slow Python loops
features_muliregression.py	Z-score normalization so features share a scale
polynomial_features.py	Feature engineering — fitting curves with x², x³
scikit_learn.py	Finally, the library version for comparison

The dot product is the heart of it:

z = np.dot(X, w) + b     # all examples, all features, at once

💡 Key insights:

• Number of weights = number of features (not examples!)
• A linear model can fit a curve if you feed it engineered features like x²
• Normalizing features makes gradient descent converge much faster

---

🎯 3. Classification — classification/

When the answer isn't a number but a category — yes/no, benign/malignant, 0 or 1.

File	What it teaches
logistical_regression.py	Why a straight line fails on categorical data
sigmoid_function.py	The sigmoid that squashes any number into (0, 1)
boundaryline.py	Drawing the decision boundary that separates classes

The sigmoid function — the star of classification:

def sigmoid(z):
    return 1 / (1 + np.exp(-z))

It turns the linear output z = w·x + b into a probability:

        z  →  -∞ ........ 0 ........ +∞
  sigmoid →   0 ......... 0.5 ........ 1
            (class 0)  (boundary)  (class 1)

💡 Key insights:

• The output is a probability, not a hard 0/1 — you threshold at 0.5
• sigmoid(z) = 0.5 exactly when z = 0, so the decision boundary is the line w·x + b = 0
• For w = [1, 1], b = -3, the boundary is the clean line x₁ = 3 - x₀

---

🧠 Concepts Collected Along The Way

A running list of the "aha" moments this project builds:

• f(x) = w·x + b is the backbone of both regression and classification
• Cost functions measure how wrong the model is
• Gradient descent is how the model learns — nudging w and b downhill
• Vectorization (np.dot) replaces slow loops and scales to many features
• Feature engineering lets a linear model bend into curves
• Normalization keeps features on a comparable scale
• The sigmoid converts scores into probabilities for classification
• Decision boundaries are just the line where the model flips its answer
• Scientific notation: e-05 ≈ "basically 0", 9.99e-01 ≈ "basically 1"

---

🚀 Running the Code

Each file is standalone — run whichever concept you want to explore:

python3 linearRegression/gradient_descent.py
python3 classification/sigmoid_function.py

Most scripts print details to the console and then open a Matplotlib plot.

🛠️ Requirements

• Python 3.8+
• NumPy
• Matplotlib
• scikit-learn (only for the comparison file)

pip install numpy matplotlib scikit-learn

📚 Credits

Inspired by Andrew Ng's Machine Learning Specialization (DeepLearning.AI), reimplemented from scratch for deeper understanding.

📄 License

Released under the MIT License.