ST7 EI Aneo

To launch the program now, use :

mpiexec /np 12 python main2.py "--infile" deap_test.txt "--verbose"

For Linux

mpirun -np 12 python main2.py ""--infile"" deap_test.txt "--verbose"

--infile Path to the files describing the problems. One file per problem

--indir Run all problems inside the directory

--verbose To see the training progress

--loops Number of runs

For the first parallelisation method (one algorithm/multiple processor), use

main2.py

For the second parallelisation method (multiple algorithms/multiple processors), change for

main_sequential.py

in the command

Todo

• meilleur algorithme d'affectations des tâches à partir de l'ordre topologique ?

• scalability issue and parallelisation ideas

• calculer les valeurs moyennes + confidence interval => hyperparameter better than another one if better values and no overlapping

parallelisation :

• get a good idea of time repartition between tasks to reduce/parallelize to improve the performance

• reduce the cardinality or complexity of the problem space if possible

Report

• Final version of the 2 reports : one less than 2 pages, short, the other one longer => temporary report on Wednesday, and the final one by Friday.
• Need to let the algorithm converge to a local minimum for the report