Types of Machine Learning

Machine Learning has three main categories. Each solves a different kind of problem.

---

1. Supervised Learning

You have answers (labels) for your training data.

The model learns the relationship between inputs and outputs.

Training Data:
    Input: house_size = 1500 sqft → Output: price = $280,000 ✓
    Input: house_size = 2000 sqft → Output: price = $350,000 ✓
    Input: house_size = 800 sqft  → Output: price = $150,000 ✓

After Training:
    Input: house_size = 1800 sqft → Output: price = ???
    Model predicts: $315,000

Two Sub-Types:

Regression — Predicting Numbers

Output is a continuous value (price, temperature, score).

from neurogebra import Expression
from neurogebra.core.trainer import Trainer
import numpy as np

# REGRESSION: Predict a number
model = Expression("linear", "m*x + b",
                   params={"m": 0.0, "b": 0.0},
                   trainable_params=["m", "b"])

X = np.array([1, 2, 3, 4, 5], dtype=float)
y = np.array([2.1, 3.9, 6.1, 7.8, 10.2], dtype=float)

trainer = Trainer(model, learning_rate=0.01)
trainer.fit(X, y, epochs=200)

Use Cases	Loss Functions	Output
House prices	MSE, MAE	Continuous number
Stock prediction	Huber	Continuous number
Temperature forecast	MSE	Continuous number

Classification — Predicting Categories

Output is a class/category (spam/not-spam, cat/dog/bird).

# The model outputs a probability using Sigmoid
from neurogebra import MathForge

forge = MathForge()
sigmoid = forge.get("sigmoid")

# Output > 0.5 → Class 1 (spam)
# Output < 0.5 → Class 0 (not spam)
print(sigmoid.eval(x=2.0))   # 0.88 → "It's spam"
print(sigmoid.eval(x=-1.5))  # 0.18 → "Not spam"

Use Cases	Loss Functions	Output
Spam detection	Binary Cross-Entropy	0 or 1
Image recognition	Cross-Entropy	Class label
Disease diagnosis	Binary Cross-Entropy	Yes or No

---

2. Unsupervised Learning

You have NO answers (labels). The model finds patterns on its own.

Data (no labels):
    [1.2, 1.3], [1.1, 1.4], [5.5, 5.6], [5.2, 5.8], [9.0, 9.1]

Model finds:
    Cluster A: [1.2, 1.3], [1.1, 1.4]         (small values)
    Cluster B: [5.5, 5.6], [5.2, 5.8]         (medium values)
    Cluster C: [9.0, 9.1]                      (large values)

Use Cases	Algorithms
Customer segmentation	K-Means clustering
Anomaly detection	Autoencoders
Data compression	PCA
Recommendation	Collaborative filtering

---

3. Reinforcement Learning

The model learns by trial and error, receiving rewards or penalties.

Environment: Video game
Agent: The AI player
Actions: Move left, right, jump
Rewards: +10 for collecting coin, -100 for falling

The agent tries random actions, learns which ones lead to rewards,
and eventually masters the game.

Use Cases	Examples
Game playing	AlphaGo, Atari games
Robotics	Walking, grasping objects
Self-driving	Navigation decisions

---

Comparison Table

Type	Has Labels?	Goal	Neurogebra Focus
Supervised	Yes	Predict from labeled data	✅ Full support
Unsupervised	No	Find patterns	Partial
Reinforcement	Rewards	Maximize reward	Future feature

Neurogebra focuses on Supervised Learning

This is where you'll spend most of your ML journey. Neurogebra excels at making supervised learning understandable — you can see every loss function, every gradient, every optimization step.

---

The ML Problem Decision Tree

What kind of output do you need?
│
├── A number (price, score, amount) → REGRESSION
│   └── Neurogebra: use MSE/MAE loss, linear output
│
├── A category (yes/no) → BINARY CLASSIFICATION
│   └── Neurogebra: use binary_crossentropy, sigmoid output
│
├── A category (cat/dog/bird) → MULTI-CLASS CLASSIFICATION
│   └── Neurogebra: use cross_entropy, softmax output
│
├── Groups in data (no labels) → CLUSTERING
│   └── Use: K-Means, DBSCAN
│
└── Sequential decisions → REINFORCEMENT LEARNING
    └── Use: Q-Learning, Policy Gradient

---

Neurogebra Makes It Clear

from neurogebra import MathForge

forge = MathForge()

# See all loss functions — pick the right one for your task
losses = forge.list_all(category="loss")
print(losses)

# Each one explains when to use it
for name in ["mse", "mae", "binary_crossentropy", "hinge"]:
    expr = forge.get(name)
    print(f"\n{name}:")
    print(f"  {expr.metadata.get('description', '')}")
    print(f"  Best for: {expr.metadata.get('usage', '')}")

---

Next: The ML Workflow →