Hello, I would like to remove specific datapoints during training, I would need to keep track of the ‘global’ index of a particular image, remove and have a new loader with the new set (after removal) set. What is the best way to do this?
I think one easy approach would be to
- store the indices of all data samples, which should be removed in a
list
during training - remove these indices from the complete list of indices to create the “valid indices”
- wrap the
dataset
in aSubset
with the new “valid” indices, - and recreate the
DataLoader
after the epoch.