Artificial generated data: (X_1, X_2) are uniformly distributed on 
[-1,1]x[-1,1]. Y is generated from Re(z^8) or

Y = X_1^8 - 28 X_1^6 X_2^2 + 70 X_1^4 X_2^4 - 28 X_1^2 X_2^6 + X_2^8

A similar dataset was used in the paper of Donoho and Johnstone,
Projection-Based Approximation and a Duality with Kernel Methods,
The Annals of Statistics 1989, pp.58-106
