Whether PCE can be improved by optimizing hyperparameters through optimization algorithms?

Dear Uqlab Team:
Thank you very much for your outstanding contribution, uqlab has played an unparalleled role in my research.
Here I have two questions to ask:
1. Whether PCE can be improved?such as the hyperparameters in the LAR method are optimized by an optimization algorithm, the type is the same as that of ANN being improved.
2. Is the number of DoEs related to the dimension of the input parameters?in other word, how to determine the right size of DoE?
Beacuse in my research I found that even small sizes (sometimes 10 or 20) can build reliable surrogate models.