Problem

CERN employs ten times more engineers and technicians than research physicists.

Do you know why?

– https://home.cern/science/engineering

Participating

1. register
2. template code & training data
3. submitting code
4. standardised & secure GPU-enabled continuous integration (CI)
5. live leaderboards
6. test data

Participating

1. register
2. template code & training data
3. submitting code
4. standardised & secure GPU-enabled continuous integration (CI)
5. live leaderboards
6. test data

Registration

Template Code

Repository

Requires SIRF, CIL, CUDA
e.g. via Docker synerbi/sirf:latest-gpu

Template Code

Layout

• (required) main.py

  from cil.optimisation.algorithms import Algorithm
  class Submission(Algorithm):
      ... # e.g. BSREM, ISTA, OSEM
  submission_callbacks = [] # per-iteration

• apt.txt/environment.yml/requirements.txt: passed to apt/conda/pip install

Submitting Code

Private mirror

• https://github.com/SyneRBI/PETRIC-{TEAM_NAME}

Continuous Integration

GitHub (native) runners

• can run basic (CPU-only) checks

Continuous Integration

GitHub self-hosted runners

Acquire or rent a machine (AWS, GCP, Azure, etc…)

• GPU (NVIDIA A100 PCIe 40GB GPU)
• CPU 32-Core
  • 100GB RAM
• sequential queue

Continuous Integration

Security

GitHub docs^1,2 TL;DR:

don’t do this because we can’t be bothered to document how to make it secure

Cloud Services

Public domain

• public IP (130.246.81.43)
• (sub)domain name (*.tomography.stfc.ac.uk)
• docker services & reverse proxy with TLS (https://)

Data

https://petric.tomography.stfc.ac.uk/data

Cloud Services

Leaderboards

During competition

Results

Fin

Want help running your own competition?

Want to compete next year?

github/casperdcl