ml
Bias-Variance Tradeoff
Training vs validation error curves — optimal complexity finder, underfitting/overfitting zones, stacked error decomposition, bulls-eye diagram.
Input
Training vs Validation Error
—— training—— validation--- optimal
Error Decomposition per Complexity
■ irreducible■ bias²■ variance
Bias-Variance Bulls-Eye (conceptual)
Analysis
Complexity levels8
Optimal complexity5
Min validation error0.36
Irreducible error0.13
Underfitting levels2
Overfitting levels5
Per-Complexity Table
cTrainValGapZone
10.850.880.03underfit
20.60.620.02underfit
30.420.450.03
40.30.380.08overfit
50.220.360.14overfit
60.170.40.23overfit
70.150.480.33overfit
80.130.60.47overfit
Bias² ≈ train error − floor · Variance ≈ val − train gap
Total error ≈ Bias² + Variance + Irreducible
Total error ≈ Bias² + Variance + Irreducible