フリー問題

問題文

Which quantity must be fixed before the training and evaluation subsets are separated, so that the separation does not have to be redone?

選択肢

1. The compression level of the intermediate files, since the files written before the separation have to stay readable by both of the later subsets and by the training job.
2. The number of components kept by the reduction step, since the width of the feature matrix has to be settled before any rows are set aside for evaluation.
3. The value used to fill missing numeric entries, because that value has to be computed over all the rows before the split so that both subsets receive the same treatment and the evaluation subset is not handled differently from the training subset in any respect.
4. The definition of the unit that must not be split across the two subsets, such as the customer when several rows belong to the same customer.