Compare OGER/Ontorunner vs Nemo-Serve/SAPBert vs Nemo-Serve/QuickUMLS on split_11

[gaurav]

Needed:

• split_11 annotations from OGER/Ontorunner
• split_11 annotations from Nemo-Serve/SAPBert
• split_11 annotations from Nemo-Serve/QuickUMLS
• Tools to compare them
  • You can see current tools at Move medtype-benchmarks comparison tools into its own repository #1

[gaurav]

We now have Ontorunner output for split_11 available on Hatteras at /projects/ner/split_11_results/ontorunner/split_11-results.jsonl.

• Number of results: 488,850 (the same number of results as split_11.txt)
• Range: 0-34640 denotations per abstract (only one abstract has zero denotations)
• Median: 207, mean: 225.62
• 25th percentile: 139, 75th percentile: 287

I'm working on documenting how I came up with these results (RENCI-NER/oger-benchmarks#2). As part of that work, I'd like to create a Docker image for OGER/Ontorunner and re-run split_11 on Docker so we have a measure of how quickly Ontorunner can process input text.