What is the best way to correct an annotation?

1

Hello,
I have exported the annotations using db-out recipe.
The person who created the annotations made mistakes, I must fix them before training. What is the best way to fix an annotation using prodigy?
Should i pass the input_hash/task_hash somewhere to load the wrong annotation and then correct/save it ?

Thank you

2

Datasets in Prodigy are “append only”. One of the reasons behind this is that it ensures you never accidentally overwrite or lose data, and always have a record of the original annotations.

So the most straightforward way to correct or review your annotations is to load the exported JSONL file back into Prodigy as the input data. The input and output data have the same format – so you can always load an annotated set back in as the stream. If the corrections are simple, you could also just make them in the JSONL file and then import it to a new dataset using db-in. Once you’re happy and verified that everything is correct, you can then delete the old set. Or keep it, just in case.

1 Like