mdwyman / surrogatebl

Surrogate Beamline -- training NN to emulate Shadow simulations of Xray beamlines

Surrogate Beamline

Code

• beamline_29.py
  • shadow implmentation fo 29ID (don't recall which branch)
• torchTools.py
  • homebrew tools to hasten NN build/training/testing

Data files

• IEX_100k_04w.pkl
  • Pickle of Shadow3 simulated data from IEX_OO_Shadow_tests.ipynb
• IEX_100k_04w_scaling.pkl
  • Pickle of feature scaling used in IEX_surrogate_NN.ipynb
• IEX_100k_04w_NN_results.pkl
  • Pickle of models and results from IEX_surrogate_NN.ipynb

Notebooks

• IEX_OO_Shadowtests.ipynb
  • Testing out beamline_29.py
  • playing with ranges of each DOF
  • build data set for surrogate
• IEX_surrogate_NN.ipynb
  • train and test NN on shadow generated data
• IEX_shadow_vs_surrogate.ipynb
  • test time usage of both beamlines
• GA_vs_beamline.ipynb
  • in progress
  • goal to test out simple genetic algorithm to find optimal beamline configuration

Upcoming

• Gaussian process implementation (variation of MG-GPO)
• Bluesky version of beamline
  • Either adapt beamline_29.py or surrogate model
  • Construct plan that can implement optimization technqiue
    • GA
    • MP-GPO
    • ???