search

texttechnologylab / DockerUnifiedUIMAInterface

A UIMA-based tool for the scaled, uniform, distributed, platform-independent and easily reusable use of Natural Language Processing (NLP) methods using Docker.
https://texttechnologylab.github.io/DockerUnifiedUIMAInterface/
GNU Affero General Public License v3.0
11 stars 10 forks source link

Duplication of results when using "composer.run" on single CAS #34

Open Myyyvothrr opened 1 year ago

Myyyvothrr commented 1 year ago

There seems to be a problem if we process multiple single CAS objects by manually "run"ing the composer. The results of the service are duplicated in the results, indicating that some intermediate CAS ist not "reset". This works fine if a collection reader is used however, and it also worked in earlier versions (a6852c774d70208140a11691c76a826068ea8a77 is the last confirmed). For testing, running the complete test suite of the sentiment tool (https://github.com/texttechnologylab/TTLab-UIMA/blob/main/textimager-uima-transformers-sentiment/src/test/java/org/hucompute/textimager/uima/transformers/sentiment/CardiffnlpTwitterXlmRobertaBaseSentimentTest.java) should fail, while running the unit tests manually one by one every test should succeed.

Myyyvothrr commented 1 year ago

Should be fixed in PR https://github.com/texttechnologylab/DockerUnifiedUIMAInterface/pull/35

@alleonhardt @abrami Could you check this change does not introduce unexpected side effects?