A question about NEP descriptor normalization

[lhycms]

I found that NEP descriptors have a normalization factor called q_scaler, which is computed using a find_max_min() kernel function. However, it is not practical to compute the NEP descriptors for the entire dataset every step just to find the global maximum and minimum values, as this would be too slow.

I would like to ask:

1. Does find_max_min() only compute the maximum and minimum values of the descriptors within the current batch?
2. If the assumption in question 1 is correct, does it mean that the final q_scaler saved in nep.txt is based on the last batch? If so, wouldn't this make q_scaler unrepresentative of the entire dataset's statistical range?

I appreciate your time and assistance in answering my questions.