Some doubtful questions

Speed and Structure
1

Hi, I would like to ask the following two questions:

  1. Are the maxand min values of the speed data in the test set consistent with those in the training set? (I don’t know if this question can be answered). This involves the normalization method
  2. g5.12xlarge has a total of 4x24=96GB of GPU memory. To ensure reproducibility on g5.12xlarge, do we need to limit the amount of GPU memory used by the training model to no more than 96GB?
1 Like
2

Good question! All of the training data(including extended) seems bound into a min and max values for the geological formation. Important to clarify incase there’s a distribution shift, which would seem strange considering the holdout is expected to be from same distribution as public test set.