✂️

Train/Test Split

From: Train/Test SplitView in lesson →

💡

How do you know if your model actually works? You can't test it on the same data it learned from—that's like letting a student see the exam answers beforehand. This tool demonstrates the critical practice of splitting data to get an honest measure of model performance.

Validation Data: 10%(auto-calculated)
Training
0
Test
0
Validation
0

Understanding the Splits

  • Training (70%): Model learns from this data
  • Test (20%): Final evaluation - NEVER touch during development
  • Validation (10%): Tune hyperparameters and compare models

Experiment as much as you like; there is no progress to lose.