Lensa ML
Bias-Variance Tradeoff
Step 1 of 6
What is Bias?
Systematic underfitting
Polynomial Degree2
DEGREE
2
BIAS²
0.488
TRAIN ERR
0.206
High bias — the model is too simple to capture the sine pattern.
Step 1 of 6: What is Bias?
Systematic underfitting
Polynomial Degree2
DEGREE
2
BIAS²
0.488
TRAIN ERR
0.206
High bias — the model is too simple to capture the sine pattern.
Step 2 of 6: What is Variance?
Sensitivity to training data
Polynomial Degree2
DEGREE
2
VARIANCE
0.043
SPREAD
0.038
Low variance — all five fits agree closely regardless of training data.
Step 3 of 6: The U-Shaped Curve
Error = Bias² + Variance + Noise
Model Complexity4
BIAS²
0.198
VARIANCE
0.126
NOISE
0.09
TOTAL
0.414
The sweet spot minimizes total error — the bottom of the U-curve.
Step 4 of 6: Underfitting Zone
High bias, low variance
Complexity1
BIAS²
0.765
VARIANCE
0.025
TOTAL ERR
0.880
DIAGNOSIS
UNDERFIT
More data won't help here — you need a more flexible model. The red lines show how far the predictions miss.
Step 5 of 6: Overfitting Zone
Low bias, high variance
Complexity12
BIAS²
0.010
VARIANCE
1.204
TOTAL ERR
1.304
DIAGNOSIS
OVERFIT
Extreme overfitting — the model memorizes noise. More data CAN help here, or use regularization to reduce variance.
Step 6 of 6: The Sweet Spot
Balancing complexity
Model Complexity5
OPTIMAL
4.6
TEST ERR
0.409
TRAIN-TEST GAP
0.250
You found the sweet spot! This complexity minimizes expected test error.