| Option | Default | Acceptable Values | Acceptable Types | Description |
|---|---|---|---|---|
| batch_size | 1000 | N/A | ['int'] | Number of samples to distribute among the processors at a time when run_parallel is True. This should be limited when the memory required to store the batch size of samples grows too large. |
| debug_print | [] | ['desvars', 'nl_cons', 'ln_cons', 'objs', 'totals'] | ['list'] | List of what type of Driver variables to print at each iteration. |
| invalid_desvar_behavior | warn | ['warn', 'raise', 'ignore'] | N/A | Behavior of driver if the initial value of a design variable exceeds its bounds. The default value may beset using the `OPENMDAO_INVALID_DESVAR_BEHAVIOR` environment variable to one of the valid options. |
| procs_per_model | 1 | N/A | ['int'] | Number of processors to give each model under MPI. |
| run_parallel | False | [True, False] | ['bool'] | Set to True to execute samples in parallel. |
Note
This feature requires MPI, and may not be able to be run on Colab or Binder.
The following script is executed using mpiexec -n 4 mpi_script_0.py:
import openmdao.api as om\n",
"from openmdao.test_suite.components.paraboloid import Paraboloid\n",
"from openmdao.utils.mpi import MPI\n",
"\n",
"samples_3x3 = [\n",
" {'x': {'val': 0.}, 'y': {'val': 0.}},\n",
" {'x': {'val': .5}, 'y': {'val': 0.}},\n",
" {'x': {'val': 1.}, 'y': {'val': 0.}},\n",
"\n",
" {'x': {'val': 0.}, 'y': {'val': 0.5}},\n",
" {'x': {'val': .5}, 'y': {'val': 0.5}},\n",
" {'x': {'val': 1.}, 'y': {'val': 0.5}},\n",
"\n",
" {'x': {'val': 0.}, 'y': {'val': 1.}},\n",
" {'x': {'val': .5}, 'y': {'val': 1.}},\n",
" {'x': {'val': 1.}, 'y': {'val': 1.}},\n",
"]\n",
"\n",
"rank = MPI.COMM_WORLD.rank\n",
"\n",
"prob = om.Problem()\n",
"\n",
"prob.model.add_subsystem('comp', Paraboloid(), promotes=['*'])\n",
"\n",
"prob.driver = om.AnalysisDriver(samples=samples_3x3, run_parallel=True)\n",
"prob.driver.add_recorder(om.SqliteRecorder('analysis_driver.sql'))\n",
"\n",
"prob.driver.add_response('f_xy', units=None, indices=[0])\n",
"\n",
"prob.setup()\n",
"prob.run_driver()\n",
"prob.cleanup()\n",
"\n",
"cr = om.CaseReader(str(prob.get_outputs_dir() / f'analysis_driver.sql_{rank}'))\n",
"cases = cr.get_cases(source='driver')\n",
"\n",
"print(f'Recorded {len(cases)} cases on proc {rank}.')\n",
"\n",
"print(f'{"x":^6} {"y":^6} {"f_xy":^6}')\n",
"for case in cases:\n",
" x = case.get_val('x')[0]\n",
" y = case.get_val('y')[0]\n",
" f_xy = case.get_val('f_xy')[0]\n",
" print(f'{x:6.3f} {y:6.3f} {f_xy:6.3f}')\n",
"\n",
"from openmdao.utils.assert_utils import assert_near_equal\n",
"import numpy as np\n",
"\n",
"assert(len(cases) == 3 if rank == 0 else 2)\n",
"# arrays in expected are [x, y, f_xy]\n",
"if rank == 0:\n",
" expected = [[0.0, 0.0, 22.0], [0.5, 0.5, 23.75], [1.0, 1.0, 27.00]]\n",
"elif rank == 1:\n",
" expected = [[0.5, 0.0, 19.25], [1.0, 0.5, 21.75]]\n",
"elif rank == 2:\n",
" expected = [[1.0, 0.0, 17.00], [0.0, 1.0, 31.00]]\n",
"else:\n",
" expected = [[0.0, 0.5, 26.25], [0.5, 1.0, 28.75]]\n",
"\n",
"for i, case in enumerate(cases):\n",
" x = case.get_val('x')[0]\n",
" y = case.get_val('y')[0]\n",
" f_xy = case.get_val('f_xy')[0]\n",
" assert_near_equal([x, y, f_xy], expected[i])\n",
"Note
This feature requires MPI, and may not be able to be run on Colab or Binder.
The following script is executed using mpiexec -n 4 mpi_script_1.py:
import numpy as np\n",
"import openmdao.api as om\n",
"\n",
"class FanInGrouped(om.Group):\n",
" """\n",
" Topology where two components in a Group feed a single component\n",
" outside of that Group.\n",
" """\n",
"\n",
" def __init__(self):\n",
" super().__init__()\n",
"\n",
" self.set_input_defaults('x1', 1.0)\n",
" self.set_input_defaults('x2', 1.0)\n",
"\n",
" self.sub = self.add_subsystem('sub', om.ParallelGroup(),\n",
" promotes_inputs=['x1', 'x2'])\n",
"\n",
" self.sub.add_subsystem('c1', om.ExecComp(['y=-2.0*x']),\n",
" promotes_inputs=[('x', 'x1')])\n",
" self.sub.add_subsystem('c2', om.ExecComp(['y=5.0*x']),\n",
" promotes_inputs=[('x', 'x2')])\n",
"\n",
" self.add_subsystem('c3', om.ExecComp(['y=3.0*x1+7.0*x2']))\n",
"\n",
" self.connect("sub.c1.y", "c3.x1")\n",
" self.connect("sub.c2.y", "c3.x2")\n",
"\n",
"prob = om.Problem(FanInGrouped())\n",
"\n",
"# Note the absense of adding design varaibles here, compared to DOEGenerator\n",
"\n",
"prob.driver = om.AnalysisDriver(om.ProductGenerator({'x1': {'val': np.linspace(0.0, 1.0, 10)},\n",
" 'x2': {'val': np.linspace(0.0, 1.0, 10)}}))\n",
"\n",
"prob.driver.add_recorder(om.SqliteRecorder("cases.sql"))\n",
"\n",
"prob.driver.add_response('c3.y')\n",
"\n",
"# # the FanInGrouped model uses 2 processes, so we can run\n",
"# # two instances of the model at a time, each using 2 of our 4 procs\n",
"prob.driver.options['run_parallel'] = True\n",
"prob.driver.options['procs_per_model'] = procs_per_model = 2\n",
"\n",
"prob.setup()\n",
"prob.run_driver()\n",
"prob.cleanup()\n",
"\n",
"# a separate case file will be written by rank 0 of each parallel model\n",
"# (the top two global ranks)\n",
"rank = prob.comm.rank\n",
"\n",
"num_models = prob.comm.size // procs_per_model\n",
"\n",
"if rank < num_models:\n",
" filename = f'cases.sql_{rank}'\n",
"\n",
" cr = om.CaseReader(prob.get_outputs_dir() / filename)\n",
" cases = cr.list_cases('driver', out_stream=None)\n",
"\n",
" values = []\n",
" for case in cases:\n",
" outputs = cr.get_case(case).outputs\n",
" values.append((outputs['x1'].item(), outputs['x2'].item(), outputs['c3.y'].item()))\n",
"\n",
" print("\\n"+"\\n".join(["x1: %5.2f, x2: %5.2f, c3.y: %6.2f" % (x1, x2, y) for x1, x2, y in values]))\n",
"