When using spans and relations, one user reports issue with no images even though there should be text. Using the 'wrap' button the text disappears ?
Welcome to the forum @tlingren 
The relations UI is not performing well with very long documents. It is a known limitation that we still need to work through.
If your documents are very long ( 7K characters is probably the limit), then you might want to look for sensible ways of splitting your documents in the snippets. Some users also worked around it successfully using a sliding window over the long documents.
However of your documents are not particularly long, could you perhaps share one of the examples that doesn't work for you? It's probably a front-end issue, since unwrapped version is being rendered correctly.
Thank you. I didn't think size of 7k was a very large document at all.
The problem we are dealing with does require relations as well as multi token spans, so we are using rel.manual with custom labels for that.
However, our annotators need to review the end of the document as well as the beginning for the full annotation decision. Is there away to navigate between these split documents (assuming we use a split method) ? Annotations might change, based on later information, for example.
Hi @tlingren,
Just to be clear I meant (roughly) 7K characters referring to the length of a document.
Before working on the right splitting method, I'd make sure it is indeed the length of the document that causes issue. If you split a single document that currently doesn't render into snippets - do you manage to render each snippet correctly?
And to answer your question about navigating between snippets. Prodigy has a History buffer visible to annotators in the bottom left corner.
If you want to go back an reannotate an example from the history you can click on it and it will show up in the main annotation panel for edition.
The size of this history is the same asbatch_size setting (10 by default). Once the the examples are saved in the DB and they will automatically be saved every 10 examples (or whatever your batch_size is), it will no longer be possible to undo or edit.
In your case you'd have to make sure then that the batch_size is at least as big as the max number of snippets per document.
Also, if you will go ahead with the snippeting make sure to store the document ID and snippet ID in meta of each example. It will be helpful for any further merging, splitting and also for computing train/dev/test splits on document level.
