SECAR quad tuning optimizer using pygmo

Requirements:

pygmo
pandas
numpy
matplotlib
scikit-learn
(a variety of other py packages are imported, but most should be preinstalled)

the python packages can be easily installed with conda using: conda env create -f environment.yml
then activate with:
conda activate secar_moead

Additionally, change the INCLUDE path in the header of the fox files in the ./fox/ directory to point to the location of COSY.bin

Files:

COSY10.0/ :

directory holding all of the necessary scripts and binaries for executing cosy

if cosy is already installed on your system you can ignore this and simply use your version presuming you update the fox files to use the correct INCLUDE path and change COSY_DIR in python scripts

fox/ :

contains all the fox files necessary for running the optimization

files should resemble SECAR_pg_Optics.fox, SECAR_pg_Optics_DE.fox,
or SECAR_an_Optics.fox, SECAR_an_Optics_DE.fox SECAR_pg_Optics.fox specifies the optics of just the recoils through SECAR and prints out the dimensions at each element in addition to the dimensions for the objectives in this case that is the X width at each FP1, 2, 3, and the beam spot size on the DSSD SECAR_pg_Optics_DE.fox specifies the optics of the unreacted beam (or whatever contaminant you'd like to separate) through SECAR and prints out the location of the contaminant for each objective in this case the separation at FP1, 2, 3

change fox_name in config.json to match the base name of your desired fox file (i.e. SECAR_pg_Optics)

as above, be sure to update the INCLUDE path in the header of the fox files to point to the location of COSY.bin

most importantly, any new fox files need to print out X and Y dimensions at every element of the beamline see cosy.py header, specifically the variables magnet_names and magnet_dims for the elements, or else use SECAR_pg_Optics.fox as a guide

both the files must also print out the desired objective values at their locations, for SECAR_pg_Optics.fox and SECAR_pg_Optics_DE.fox that means the recoil widths and contaminant separations at FP1, 2, 3 and the recoil beam spot size at the DSSD

the MaxBeamWidth is then calculated by taking the largest width of the recoils relative to the width of each element see cosy.py for specifics

py/ :

analyze_db.py

defines a method to read the hdf5 db and write only values which are on the pareto front and are less than max_obj to a separate db also runs a kmeans-clustering algorithm on the db to generate clusters of points

config.json

specifies most of the variables needed for running the full process comments don't exist in json so additional keys like fNominalan and fNominalpg are unused they are simply there to hold the values

cosy.py

defines cosyrun(input) that runs a COSY simulation given input magnet settings and returns a set of objective values assumes cosy is installed and can be run with: cosy foxFile.fox all the interfacing between cosy and python is done here, so be sure to check that, given a specified reaction type and fNominal values, ./cosy.py returns values which are == 1 or only off by a precision < 1e-7 or so

optimize.py

defines main() which implements the pygmo archipelago and runs the optimization evolution (note that the archipelago can be a single island)

pca.py

runs a pca on the specified results and then generates a number of random points using those components

pipeline.py

loads in all the necessary python scripts and runs them to generate and plot results

problem.py

defines a pygmo User-Defined Problem which calls cosyrun in its fitness evaluation

utils.py

defines various useful functions, most notably providing the wrapper to read in the config.json file

Drawing scripts :

Several different draw*.py scripts exist, in various states of clarity. I will continue to try to clean and comment these but they are less important than the analysis scripts

draw.py

defines a method for plotting the results requires an argument of the file to be analyzed e.g. run with:
./draw.py ../output/outputFileName.csv

sh/ :

all the bash scripts for submitting jobs to fireside are here

analysis_job.sh

launches a job for pca.py, generating the pca points

slurmfiles/ :

try to write out the sbatch logs to this file so they can be easily stored and cleaned as necessary

submit_job.sh

launches a job for optimize.py

Deprecated scripts:

results are no longer stored to csv, and so making them into a pandas df is trivial

make_db.py

defines a method for constructing a pandas hdf5 database from the csv results db to be made is specified within the code

pipeline.sh

basic pipelining done in bash, updated to the pipeline.py script

Notes:

assuming python3 is installed with conda and
which python3
points to the correct version all scripts can simply be made executable and called with:
./[script name] [args]