Course contentsShow
← Machine Learning and Deep Learning
Lesson 525 of 3,53812. Model Validation and TuningFree lesson

Model Complexity Curves

Visualizing the bias-variance tradeoff by varying model complexity and observing error trends.

Model Complexity Curves

You'll learn how to visualize the bias-variance tradeoff by plotting model complexity against training and validation errors.

What Are Model Complexity Curves?

A model complexity curve is a plot that shows how both training error and validation error change as you increase the complexity of your model. Unlike validation curves (which vary a single hyperparameter) or learning curves (which vary dataset size), complexity curves give you the big picture of the bias-variance tradeoff.

Think of it like tuning a guitar string: too loose (simple model) and it won't play the right note (high bias), too tight (complex model) and it might snap or sound harsh (high variance). The complexity curve helps you find the sweet spot.

How It Works

As model complexity increases:

  • Training error decreases: More complex models fit the training data better and better
  • Validation error follows a U-shape: It decreases initially (model learns true patterns), then increases again (model starts overfitting)

The optimal complexity lies at the bottom of the validation curve's U-shape, where you balance underfitting and overfitting.

What You See

  • Left side (low complexity): Both errors are high → underfitting/high bias
  • Sweet spot (middle): Validation error minimized → just right
  • Right side (high complexity): Large gap between training and validation error → overfitting/high variance

This visualization builds on everything you've learned about high bias, high variance, and validation curves—bringing it all together into one diagnostic tool.

Key Takeaway: Model complexity curves reveal the entire bias-variance spectrum at once, helping you choose model complexity that generalizes best to new data, not just memorizes training examples.