📌 多項式回帰と過学習

次数 d

3

点の数

60

ノイズ σ

0.10

学習データ比

70%

Train MSE: -

Test MSE: -

Degree vs Error

Ground truth: y = sin(2πx) + 0.3x + 0.1
Higher-degree polynomials often reduce training error but may increase test error (overfitting).

How to read the charts

Left: points Train=blue, Test=red; gray dashed = ground truth; blue line = fitted curve.
Top metrics: current Train/Test MSE.
Right: Train and Test vs degree d. The degree where Test is minimal is a good choice.
As degree increases, the curve gets more complex; Train typically decreases, while Test rising indicates overfitting.