Deselect all incorrect labels in NER

Hello,

I need to deselect all highlighted labels to correct a NER model (highlighted blocks of text that should not be highlighted) instead of having to click each label individually. Any help would be greatly appreciated!!
(see image attached)

Hi! There's no built-in feature in the UI at the moment, but here's a little hack you could use:

  • In your prodigy.json, set "javascript": "console.log('JS enabled!')" or similar (anything, really – it just needs to be set so Prodigy enables custom JavaScript).
  • Open your browser's developer console and type in: window.prodigy.update({'spans': []}). This will update the current task and will remove all selected "spans".

Btw, are these predictions produced by your model, or results of some previous process? If you end up with all tokens highlighted, this indicates that something went wrong and it's probably easier to fix in a preprocessing step, or by retraining your model.

1 Like