readme update for memory usage of Verb SRL in pipeline (close #656); ChunkerTrain bug fixed and model updated (close #685) #694
Conversation
…ise, it's affected by the existing pretrained models.
qiangning
changed the title
|
Note: I'm working on two issues and they got merged into this single PR. The first commit of The other three commits are to close #685. I fixed the bug of ChunkerTrain that it wrongly loads pretrained models when training a new one. I then retrained the model and found that the performance got slightly improved. So I deployed the new model to here and changed the code accordingly. |
|
@qiangning looks like CI build is failing -- please take a look. |
|
It should be that the mvn test failed. I'm working on it. |
|
@qiangning CI test is failing because chunker output has changed, and the test diffs it with a reference output. Please update the test files and this should be good to go. |
|
Thanks. Yes, I found it. Plan to work on it tonight. |
|
All the tests have passed. If no objections, I'll merge it sometime late tonight. |
There was a problem hiding this comment.
It looks like you deleted/renamed some files. This is fine, but I want to retain the word "ref" or "reference" in the output file that is retained in the repository, to distinguish it from test time output. Therefore,
the two files to retain are testIn.txt and testRefOut.txt. These can go in src/test/resources/, as they should be part of the version-controlled repository.
The diff and out files should not go in src/test/resources/ as they are generated by the user running the test and are temporary.
After investigation of the memory usage of Verb SRL in our pipeline, I suggest that we close #656 with a few notes added in README.