#structural-error

The true regression function is $f(x) = x_1^3 x_2^5 - 10x_1x_2 + 7x_1^2 + 5x_2 - 3$ on the square $[-1, 1]^2$, observed with Gaussian noise of variance 1. You fit four least-squares models that use all polynomial features up to order 1, 2, 8 and 10, this time with a million training points. Call the error of the best model in a class its structural error, and the extra error from estimating its parameters from data its estimation error. Which statements are true? Select all that apply.