首先这个问题先要确定 xgboost 是决策树还是线性回归。 如果是线性回归，mu 一定要严格的满足正态分布。

如果是决策树。

Yes, algorithms based on decision trees are completely insensitive to the specific values of predictors, they react only to their order. It means that you don’t have to worry about “non-normality” of your predictors. Moreover, you can apply any monotonic transformation to your data, if you want - it will not change predictions of decision trees at all! stats

这是针对 x 变量的。

As each leaf has a classification decision that is in fact a discrete distribution, one that has 100% for the class it indicates and 0 for all other classes. You could use the training set to generate a distribution for all inner nodes if you want, as well. Stack Overflow

也就是说如果是决策树，那么这里不需要假设 y 的分布，实际上 y 一直是一个离散变量，符合自己的样本分布。 他的切分是靠基尼残差，而且是在每个切分点进行，正态分布这个假设从始至终没有限制。