Currently, the databunch is setup such that you split the data into train and val datasets, and it’s a uniform split. Is there anyway to modify the train or val dataset, such that certain data points will be present in train, and not in val or the other way around?
Right now, I’m using U-Net to make predictions. I need to remove a specific data points from the training set, but I need to include that inside the val set. Those specific data points are flagged in my pandas dataframe, but I’m trying to figure out how to use that with the databunch.
Ideally, what I want to is after the train vs val split, is that I have a way to go into the train databunch, and remove all the datapoints I don’t want.