Train-Test Split
Domain: Data & Datasets
🎯 Motivation
Dividing data into training set (80%) and test set (20%). Train on one, test on the other. Prevents evaluating on training data.
📋 Concrete Example
Split: 10,000 examples. Train: 8,000 (learn from these). Test: 2,000 (evaluate on these). The model never sees test data during training. Honest evaluation.
🔗 Analogy
A teacher’s practice exam vs real exam - practice (training) questions are studied, real (test) questions are new. The split ensures honest evaluation.