Following from #129 , we have forced train/val splits to be the same for training of all posteriors in the ensemble. However, we use random assignment to specify these splits, whereas we could easily load them from file. This might be a desired feature for advanced users, or those wanting to compare many ensemble configurations.
Following from #129 , we have forced train/val splits to be the same for training of all posteriors in the ensemble. However, we use random assignment to specify these splits, whereas we could easily load them from file. This might be a desired feature for advanced users, or those wanting to compare many ensemble configurations.