Instead of randomly partitioning IAMLines into new paragraphs and saving the result to disk, we now generate a new paragraph on the fly inside the DataLoader from lines that are saved to disk.
Specifically, we add a PyTorch Dataset that combines lines together into paragraphs at indexing time, using the first provided index as a seed.
Instead of randomly partitioning IAMLines into new paragraphs and saving the result to disk, we now generate a new paragraph on the fly inside the
DataLoader
from lines that are saved to disk.Specifically, we add a PyTorch
Dataset
that combines lines together into paragraphs at indexing time, using the first provided index as a seed.supercedes #40