K****n 发帖数: 5970 | 1 如果 model Y=X
那么X在全域内分布很不均匀,要紧吗? 会体现在哪些统计量上哪? 是不是prediction
error在数据少的区域会比较大? 对于model validation有什么影响吗? 有什么方法解
决吗?
谢谢! | x**g 发帖数: 807 | | j*****e 发帖数: 182 | 3 Actually, if you are sure that X and Y are linearly related, the optimal way
to estimate the model parameter is to measure Y at X=lower boundary and X=
upper boundary (The proportions of X at low and upper boundary are equal).
This is a classical result in optimal experimental design.
However, if the relationship between Y and X are not linear, you will be in
deep trouble for having a design like this. There are ways to allocate the
design points for quadratic models. Please refer to a response |
|