Skip to content

Latest commit

 

History

History
49 lines (31 loc) · 2.78 KB

README_parameter_sweep.md

File metadata and controls

49 lines (31 loc) · 2.78 KB
Raw

Running parameter sweeps

  1. Open parameter_sweep.py

  2. At the top of the file, set the parameters to be whatever you need. That is edit the contents of these arrays

    p_infect_given_contact_list = [round(0.02+0.001*i, 3) for i in range(12)]
mortality_multiplier_list = [1, 2, 3, 4, 5]
start_date_list = [10, 13, 16, 19, 22, 25]

  3. Run the following command: python3 parameter_sweep.py N_SIMS_PER_COMBO N_SIMS_PER_JOB --do_the_math

    • Where:

      • N_SIMS_PER_COMBO = number of simulations you want per parameter combination
      • N_SIMS_PER_JOB = number of simulations you want to run per slurm job

    • Note: the larger N_SIMS_PER_JOB, the more time and memory it will take. But the smaller N_SIMS_PER_JOB, the less time and memory it will take, but more jobs and CPUs.

  4. The script will print out the command you need to pass to slurm in order to run the simulations. e.g.,

    python3 parameter_sweep.py 50 10 --do_the_math --sim_name lombardy_batch

    prints

    NUM COMBOS: 360
NUM INDICES: 1800

Please launch jobs indexed from 0 to 1799, e.g.,
sbatch --array=0-1799 job.parameter_sweep.sh 50 10 lombardy_batch

  5. On a slurm-capable device, run the above sbatch command. Note that results will be saved in the directory you specify as the final (4th) position argument of the sbatch command (so lombardy_batch in the above example)

  6. When things finish, you can either jump to step 8 to plot immediately, or run step 7 to combine your result files into a more compact format (will make the rest of the plot code faster)

  7. Open combine_results.py. In the first if country == 'x' block, replace the parameter lists with your own. Also replace runs = ['lombardy_327_0', 'lombardy_327_1'] with the list of directories where you saved the results of all your previous simulations of this set of parameters (so the directories you specified in step 5.) In NUM_JOBS, specify how many jobs were run per combo (N_SIMS_PER_COMBO / N_SIMS_PER_JOB) per batch. Set combined_dir = 'lombardy_327_combined' or whatever you wish as an output directory. Run python3 combine_results.py N_PROCESSES where N_PROCESSES = number of processes you want (the code is parallelized).

  8. To plot results, open parse_parameter_sweep.py

  9. In the if country == 'x' block at the top (near line 18), replace the parameter lists with your own.

  10. In the if country == 'x' block around line 78, replace runs = [] with the list of directories containing your results. Note that it can handle many directories. If you ran many batches of trials, you can specify the directories containing the results of each here and the plotting code will retrieve them. Or if you ran step 7, you only need to pass the one combined directory (this will be faster.)

  11. Run python3 parse_parameter_sweep.py