Per NDS-935, let's re-run cross-validation with just the test queries. Per Garrick's suggestion, there's no need to re-run the actual runs, just extract the results from the eval directory for the previous runs.
I've created a login for you on biocaddie.ndslabs.org (will send credentials via slack).
Data from previous runs is in /data/willis8/biocaddie/. The directory structure should be what you are familiar with.
- For all baseline models, write a script (or grep) the test queries from the eval/model/combined/short. Create new eval files containing only test data in eval/model/test/short. Re-run LOOCV process.
- Create a new results page under https://opensource.ncsa.illinois.edu/confluence/display/NDS/BioCADDIE+Results similar to your existing pages.
Let me know if any of this is unclear or if you have problems accessing the data on the biocaddie.ndslabs.org system.