def load(cls, db_path, db_type, int_features=False):
""" Creates Predictor by loading from a database
:param db_path see load_from_db
:param db_type see load_from_db
:param int_features bool indicating if int_features are present
"""
net = prepare_prediction_net(db_path, db_type)
meta = load_from_db(db_path, db_type)
parameters = GetBlobs(meta, predictor_constants.PARAMETERS_BLOB_TYPE)
return cls(net, parameters, int_features)
def plan(
cls,
net,
simulator_parameters: PolicyEvaluatorParameters,
db_type: str,
) -> str:
for name, value in simulator_parameters.global_value_inputs:
workspace.FeedBlob(name, value)
for model in simulator_parameters.value_input_models:
if model.path is not None and len(model.path) > 0:
model_net = prepare_prediction_net(model.path, db_type)
# By default, inputs are not remapped, so let's force a remapping
meta = load_from_db(model.path, db_type)
model_net_inputs = GetBlobs(
meta, predictor_constants.INPUTS_BLOB_TYPE
)
remap = {}
for inp in model_net_inputs:
remap[inp] = '{}_{}'.format(model.name, inp)
logger.debug("REMAP VALUE MODEL: {}".format(remap))
model_net, _ = core.clone_and_bind_net(
model_net,
'new_{}'.format(model.name, ),
'{}_'.format(model.name),
blob_remap=remap,
)
net.AppendNet(model_net)
policy_net = prepare_prediction_net(
simulator_parameters.propensity_net_path, db_type
)
meta = load_from_db(simulator_parameters.propensity_net_path, db_type)
policy_net_outputs = GetBlobs(
meta, predictor_constants.OUTPUTS_BLOB_TYPE
)
assert policy_net_outputs == ['ActionProbabilities']
net.AppendNet(policy_net)
return 'ActionProbabilities'