def continue_training(self, trackers: List[DialogueStateTracker],
**kwargs: Any) -> None:
if not self.is_ready():
raise AgentNotReady("Can't continue training without a policy "
"ensemble.")
self.policy_ensemble.continue_training(trackers, self.domain, **kwargs)
self._set_fingerprint()
def _ensure_agent_is_ready(self) -> None:
"""Checks that an interpreter and a tracker store are set.
Necessary before a processor can be instantiated from this agent.
Raises an exception if any argument is missing."""
if not self.is_ready():
raise AgentNotReady("Agent needs to be prepared before usage. "
"You need to set an interpreter, a policy "
"ensemble as well as a tracker store.")
def persist(self, model_path, dump_flattened_stories=False):
# type: (Text, bool) -> None
"""Persists this agent into a directory for later loading and usage."""
if not self.is_ready():
raise AgentNotReady("Can't persist without a policy ensemble.")
self._clear_model_directory(model_path)
self.policy_ensemble.persist(model_path, dump_flattened_stories)
self.domain.persist(os.path.join(model_path, "domain.yml"))
self.domain.persist_specification(model_path)
logger.info("Persisted model to '{}'"
"".format(os.path.abspath(model_path)))
def train(self,
training_trackers: List[DialogueStateTracker],
**kwargs: Any
) -> None:
"""Train the policies / policy ensemble using dialogue data from file.
Args:
training_trackers: trackers to train on
**kwargs: additional arguments passed to the underlying ML
trainer (e.g. keras parameters)
"""
if not self.is_ready():
raise AgentNotReady("Can't train without a policy ensemble.")
# deprecation tests
if kwargs.get('featurizer'):
raise Exception("Passing `featurizer` "
"to `agent.train(...)` is not supported anymore. "
"Pass appropriate featurizer directly "
"to the policy configuration instead. More info "
"https://rasa.com/docs/core/migrations.html")
if kwargs.get('epochs') or kwargs.get('max_history') or kwargs.get(
'batch_size'):
raise Exception("Passing policy configuration parameters "
"to `agent.train(...)` is not supported "
"anymore. Specify parameters directly in the "
"policy configuration instead. More info "
"https://rasa.com/docs/core/migrations.html")
if isinstance(training_trackers, str):
# the user most likely passed in a file name to load training
# data from
raise Exception("Passing a file name to `agent.train(...)` is "
"not supported anymore. Rather load the data with "
"`data = agent.load_data(file_name)` and pass it "
"to `agent.train(data)`.")
logger.debug("Agent trainer got kwargs: {}".format(kwargs))
check_domain_sanity(self.domain)
self.policy_ensemble.train(training_trackers, self.domain,
**kwargs)
self._set_fingerprint()
logger.info("Model directory {} exists and contains old "
"model files. All files will be overwritten."
"".format(model_path))
shutil.rmtree(model_path)
else:
logger.debug("Model directory {} exists, but does not contain "
"all old model files. Some files might be "
"overwritten.".format(model_path))
def persist(self,
model_path: Text,
dump_flattened_stories: bool = False) -> None:
"""Persists this agent into a directory for later loading and usage."""
if not self.is_ready():
raise AgentNotReady("Can't persist without a policy ensemble.")
self._clear_model_directory(model_path)
self.policy_ensemble.persist(model_path, dump_flattened_stories)
self.domain.persist(os.path.join(model_path, "domain.yml"))
self.domain.persist_specification(model_path)
logger.info("Persisted model to '{}'"
"".format(os.path.abspath(model_path)))
def visualize(self,
resource_name: Text,
output_file: Text,
max_history: Optional[int] = None,
nlu_training_data: Optional[Text] = None,
