Running at Analysis Facilities#

Here is some information about running the notebook at different analysis facilities (more will be added soon).

Coffea-Casa at UNL#

To use this facility, follow the instructions at this site: https://coffea-casa.readthedocs.io/en/latest/cc_user.html. This facility includes ServiceX, Triton, and will include mlflow. In the file config.yaml, ensure that the following parameters are set:

  • AF: coffea_casa

  • TRITON_URL: agc-triton-inference-server:8001

Elastic Analysis Facility at Fermilab#

To use this facility, follow the instructions at this site: https://eafjupyter.readthedocs.io/en/latest/. A Fermilab account is required to access this facility. This facility includes Triton. In the file config.yaml, ensure that the following parameters are set:

  • AF: EAF

  • TRITON_URL: triton.apps.okddev.fnal.gov:443

  • USE_SERVICEX: False

mlflow Instructions#

If you want to try using mlflow in the training pipeline on an analysis facility that does not include mlflow, you can use a local setup as described in this documentation: https://mlflow.org/docs/latest/quickstart.html. NSCA also has a good set-up if you do not want to store models locally, and you can sign up for an account here: https://wiki.ncsa.illinois.edu/display/NCSASoftware/MLFlow+at+NCSA.