2024-12-11 MT/LW

Where to go from here?

MT/LW

LW: Nice-to-haves include

• Dask + Kubernetes cluster in released version of Evaluator

• Docker image with Evaluator + everything installed

  • Useful for fitting on NRP

  • Maybe useful more broadly

    • LW: currently using this image

      • env file:

      • entrypoint script:

    • Lily’s awesome Docker automation:

      • CUDA version hardcoded because that’s what NRP (currently) uses

Testing:

• (stretch) have some sort of automated regression testing, probably run in release process and not on every commit

Scripts for running test jobs on NRP (or other platforms)?

MT

• Input scripts:

  • How to use:

    • Untar targets and have all scripts + targets/ directory at the same level

    • run-fit.sh executes run-fit.py. This script:

      • Writes out cluster-spec.yaml

      • copies server-existing.py to a runner on Kubernetes

      • Runs a fit using ForceBalance.py optimize.in

    • Connection options are in targets/phys-prop/options.json. To connect to the EvaluatorServer using a port other than 8998, edit this file and pass in a different port to run-fit.py::main

    • Physical properties to fit are in targets/phys-prop/training-set.json. To reduce the number of physical properties, edit the "properties" list. The original list of 1k+ targets is in old-training-set.json