How to interpret IMDB sentiment predictions?

Part 1 (2018)
13

Hey, I didn’t change the model structure - I just modified this line when defining TextData.from_splits from

trn_iter,val_iter = torchtext.data.BucketIterator.splits(splits, batch_size=bs)

to

trn_iter, val_iter, test_iter = torchtext.data.BucketIterator.splits(splits, batch_sizes=(bs, bs, 1))

You can see from the source code here that torchtext.data.BucketIterator.splits actually takes in a batch_sizes tuple argument that defines batch sizes for different datasets.