search

UW-xDD / blackstack

Entity extraction from PDFs with Tesseract and Machine Learning
MIT License
11 stars 5 forks source link

Should there be so many unlabelled areas? #4

Open metazool opened 6 years ago

metazool commented 6 years ago

I'm close to having a debugged version of blackstack working in docker containers on our fork of the project at https://github.com/BritishGeologicalSurvey/blackstack now.

However I'm seeing the annotator server fall over because the SQL JOIN is returning empty, the cause is that many of the areas in my set of test documents have not got labels - about 80% of them. Some of the test docs are intentionally low quality. Is this expected and is the best thing to do to limit the query to only labelled areas, or to track down the cause of no area labelling?

jczaplew commented 6 years ago

Is the join you are referring to this one - https://github.com/UW-Deepdive-Infrastructure/blackstack/blob/master/annotator/server.py#L197 ?

metazool commented 6 years ago

That's the one, it's the ON area_labels.area_id line that it fails to return data on ( without querying for the label and only up to the first join it ran ok)

I had roughly 1k labels for roughly 5k areas in the test set...

jczaplew commented 6 years ago

Interesting...that query is intended to fetch only the areas that have been labeled in order to show p values while tagging areas. Since it isn't a functional piece of the annotator it is safe to remove it if it is being problematic.

I'm assuming the labels you have for areas are stored in the area_labels table?