I know in the official doc, there is a recommendation for concatenating datasets together. However, for my training, I have different data from different datasets that are of different sizes. My network only takes in one size for each batch, so I need a way to implement dataloaders in a way that can be rotated each epoch/step.
I thought about using yield in a concatenated dataset case similar to the one mentioned in the doc, but I’m not sure if len will be called an equal number of times as get_item.