- PDTB Dataset requires a LDC license, so the data cannot be distributed by us.
- Dataset can be obtained here: https://catalog.ldc.upenn.edu/LDC2008T05
BERT Parser (https://github.com/najoungkim/pdtb3)
- Requires the PDTB dataset to be in .csv format
- Can be obtained here: https://github.com/cgpotts/pdtb2
- Setup data paths in preProcessing function in BERTParse.ipynb
- Run preProcess function
- Change data paths and settings in test_run.sh before running
- Run powerUsage.ipynb once iterations begin
- Manually running powerUsage.ipynb cell once iterations end
- Use csvCalculation.ipynb to examine results
Non-bert (Java) Parser (https://github.com/WING-NUS/pdtb-parser)
- Trained and tested on PDTB data
- Setup data paths in config.properties before preprocessing and running
- Follow the steps on the parser's repository to determine which .jars to run
- Use rapl_power_meter and nvidia-smi commands while running to determine power usage
- Can use csvCalculation.ipynb to examine results of .csv outputs of rapl_power_meter and nvidia-smi