search

biolab / orange3-text

🍊 :page_facing_up: Text Mining add-on for Orange3
Other
127 stars 84 forks source link

Corpus forgets Used Text Features when input data subset changes #1038

Open wvdvegte opened 8 months ago

wvdvegte commented 8 months ago

Describe the bug If, from the same dataset, Corpus receives a different subset, its Used Text Features change to whatever it considers as default

To Reproduce Steps to reproduce the behavior: Open Corpus bug.ows.zip. Check that in Corpus, Content is set to Used and Title to Ignored Open Select Rows and change the selection criterion. Reopen Corpus, and conclude that now both Title and Content are set to Used.

Expected behavior The selection in Corpus shouldn't change. The data originates from the same dataset.

Orange version: 3.36.2

Orange - About - version number.

Text add-on version: 1.15.0

Operating system: OS X 14.3.1

ajdapretnar commented 7 months ago

This is related to CorpusContextHandler, which computes the hash of the corpus for each update (L79). I will leave this to the experts: @lanzagar @janezd @markotoplak @VesnaT.