Running parameter optimization for an AMICI model using parpe::OptimizationApplication

This document describes the output and options of parPE executables based on parpe::OptimizationApplication.

TODO

For now see example notebook in ../examples/parpeamici/steadystate/.

Command line options

Run the created executable with the -h/--help argument.

Output explained

Optimizer iterations

[2018-08-26 12:23:02] [INF] [   0/i21r01c01s01] [o4i2] iter: 2 cost: 50339.6 time_iter: wall: 278.65s cpu: 59638.5s time_optim: wall: 1312.94s cpu: 230120s
 ^timestamp            ^loglvl ^rank/hostname ^optimizerProgress

  • Log level: one of debug (‘DBG’), information (‘INF’), warning (‘WRN’), error (‘ERR’), critical (‘CRI’)

  • Rank: MPI rank. Master is 0. For non-MPI runs this is -1.

  • Optimizer progress: Number after ‘o’ a running index of optimizer runs, number after ‘i’ is the current iteration

  • ‘iter’: current iteration

  • ‘cost’: current cost function value

  • ‘time_iter: wall:’ wall-time for this iteration, ‘cpu’ approximate CPU-time for this iteration

  • ‘time_optim’: cumulative ‘time_iter’ for the given optimizer run

Simulations during cost function evaluation

[2018-09-05 12:14:59] [DBG] [18/i20r01c01s11] [o0i0c1717] Result for 2379: -0,792136 (0) (21,6578s+)
 ^timestamp           ^loglvl ^rank/hostname ^optimizerProgress   ^messageID ^llh ^status ^time ^gradientFlag

  • see above

  • Optimizer progress: see above + Number after ‘c’ is the current condition index

  • messageID: internal use only

  • llh: Log-likelihood as provided by AMICI

  • status: AMICI status

  • time: simulation wall-time

  • gradientFlag: ‘+’ indicates simulation with gradient, ‘-’ indicates simulation without gradient

Environment variables

  • PARPE_NUM_PARALLEL_STARTS

    Setting PARPE_NUM_PARALLEL_STARTS=n will create a maximum of n threads for concurrent local optimizations during multi-start optimization. If unset, this defaults to the number of concurrent threads supported by hardware.

  • PARPE_LOG_SIMULATIONS

    Setting PARPE_LOG_SIMULATIONS=1 will cause every single AMICI simulation to be saved in the result files.

    Note: In the case of large models and many simulation conditions this can result in huge files.

  • PARPE_NO_DEBUG

    With PARPE_NO_DEBUG=1 no LOGLVL_DEBUG messages (i.e. those prefixed with [DBG]) will be printed.

  • PARPE_MAX_SIMULATIONS_PER_PACKAGE=n and PARPE_MAX_GRADIENT_SIMULATIONS_PER_PACKAGE=n

    These environment variables determine how many simulation requests will be sent to workers within one message. For simulations without sensitivities (e.g. during IpOpt line-search) PARPE_MAX_SIMULATIONS_PER_PACKAGE is used, for simulations with sensitivities PARPE_MAX_GRADIENT_SIMULATIONS_PER_PACKAGE is used. Changing these values may be useful to reduce communication overhead when working with a high number of workers. It may also be necessary to reduce this value in the case only very few simulations can be performed in parallel.

    Note: These variables have no effect in case of shared-memory (non-MPI) execution

  • PARPE_NUM_SIMULATION_TRIALS (integer) and PARPE_INTEGRATION_TOLERANCE_RELAXATION_FACTOR (float)

    In case of simulation failure, parPE try rerunning an AMICI simulation with a PARPE_INTEGRATION_TOLERANCE_RELAXATION_FACTOR-fold higher error tolerance for a total of PARPE_NUM_SIMULATION_TRIALS times (including the initial attempt).