Currently, 7net redundantly computes the statistics of dataset and retrieve shift scale for each process. It may run incorrectly if some shift scale is obtained from a regression model, if the algorithm is not deterministic.
While there is no such method exists in 7net (at least for now) and we use fixed random seed for numpy and torch at the beginning, it is still good practice to ensure it.
Currently, 7net redundantly computes the statistics of dataset and retrieve shift scale for each process. It may run incorrectly if some shift scale is obtained from a regression model, if the algorithm is not deterministic.
While there is no such method exists in 7net (at least for now) and we use fixed random seed for numpy and torch at the beginning, it is still good practice to ensure it.