Cross Modeling

Collecting Dataset

To collect a data set for a single system and app pair use the analysis/collect-dataset.py script. You can run it as

python3 collect-dataset.py --root ../data/<system_name>/<app_name>/ --output <output_csv>

# for instance
python3 collect-dataset.py --root ../data/quartz/laghos --output ../data/quartz/laghos/data.csv

This will build a csv combining the results for all the quartz laghos runs. This script will also tell you if any of the runs failed or produced other weird outputs. You can add the --clean flag to have it remove bad results directories. To combine all of the datasets use analysis/combine-datasets.py. This can be run as

python3 combine-datasets.py -v --root ../data --output ../data/data.csv

This will combine all of the datasets into a single csv file.

Backing Up

Another helper script is data-collection/backup-data.bash. It will produce a zipped tar file with all the data in it. You can also uncomment the htar line to get it to save to taped archives. Simply run bash backup-data.bash and it will produce data.bkp.tar.gz.

Workflow

To collect data I ran the following steps:

# Run collect-dataset.py for each system/app pair
cd analysis
python3 collect-dataset.py --root ... --output ...

# Combine
python3 combine-datasets.py -v --root ../data --output ../data/data.csv

# And finally backup
cd ../data-collection
bash backup-data.bash