def run_plugin_instances(self, transport, communicator=MPI.COMM_WORLD):
out_data = self.get_out_datasets()
extra_dims = self.extra_dims
repeat = np.prod(extra_dims) if extra_dims else 1
param_idx = pu.calc_param_indices(extra_dims)
out_data_dims = [len(d.get_shape()) for d in out_data]
param_dims = [range(d - len(extra_dims), d) for d in out_data_dims]
for i in range(repeat):
if repeat > 1:
self.set_parameters_this_instance(param_idx[i])
for j in range(len(out_data)):
out_data[j].get_plugin_data()\
.set_fixed_directions(param_dims[j], param_idx[i])
self.pre_process()
logging.info("%s.%s", self.__class__.__name__, 'process')
transport.process(self)
logging.info("%s.%s", self.__class__.__name__, 'barrier')
self.exp.barrier(communicator=communicator)
logging.info("%s.%s", self.__class__.__name__, 'post_process')
self.post_process()
def _run_plugin_instances(self, transport, communicator=MPI.COMM_WORLD):
""" Runs the pre_process, process and post_process methods.
If parameter tuning is required, loop over the methods and set the
correct parameters for each run. """
out_data = self.get_out_datasets()
extra_dims = self.extra_dims
repeat = np.prod(extra_dims) if extra_dims else 1
param_idx = pu.calc_param_indices(extra_dims)
out_data_dims = [len(d.get_shape()) for d in out_data]
param_dims = [range(d - len(extra_dims), d) for d in out_data_dims]
if extra_dims:
init_vars = self.__get_local_dict()
for i in range(repeat):
if extra_dims:
self.__reset_local_vars(init_vars)
self._set_parameters_this_instance(param_idx[i])
for j in range(len(out_data)):
out_data[j]._get_plugin_data()\
.set_fixed_directions(param_dims[j], param_idx[i])
logging.info("%s.%s", self.__class__.__name__, 'pre_process')
self.base_pre_process()
self.pre_process()
logging.info("%s.%s", self.__class__.__name__, 'process')
transport._process(self)
logging.info("%s.%s", self.__class__.__name__, '_barrier')
self.exp._barrier(communicator=communicator)
logging.info("%s.%s", self.__class__.__name__, 'post_process')
self.post_process()
self.base_post_process()
for j in range(len(out_data)):
out_data[j].set_shape(out_data[j].data.shape)
def _run_plugin_instances(self, transport, communicator=MPI.COMM_WORLD):
""" Runs the pre_process, process and post_process methods.
If parameter tuning is required, loop over the methods and set the
correct parameters for each run. """
out_data = self.get_out_datasets()
extra_dims = self.extra_dims
repeat = np.prod(extra_dims) if extra_dims else 1
param_idx = pu.calc_param_indices(extra_dims)
out_data_dims = [len(d.get_shape()) for d in out_data]
param_dims = [range(d - len(extra_dims), d) for d in out_data_dims]
if extra_dims:
init_vars = self.__get_local_dict()
for i in range(repeat):
if extra_dims:
self.__reset_local_vars(init_vars)
self._set_parameters_this_instance(param_idx[i])
for j in range(len(out_data)):
out_data[j]._get_plugin_data()\
.set_fixed_directions(param_dims[j], param_idx[i])
logging.info("%s.%s", self.__class__.__name__, 'pre_process')
self.base_pre_process()
self.pre_process()
logging.info("%s.%s", self.__class__.__name__, 'process')
transport._process(self)
logging.info("%s.%s", self.__class__.__name__, '_barrier')
self.exp._barrier(communicator=communicator)
logging.info("%s.%s", self.__class__.__name__, 'post_process')
self.post_process()
self.base_post_process()
for j in range(len(out_data)):
out_data[j].set_shape(out_data[j].data.shape)
