def _process_event_with_trackers(self, event, trackers):
# type: (Event, List[FeaturizedTracker]) -> TrackerResult
"""Logs an event to all trackers.
Removes trackers that create equal featurizations.
From multiple trackers that create equal featurizations
we only need to keep one. Because as we continue processing
events and story steps, all trackers that created the
same featurization once will do so in the future (as we
feed the same events to all trackers)."""
# collected trackers that created different featurizations
unique_trackers = []
featurizations = set()
# collected training data
features = []
labels = []
for tracker in trackers:
if isinstance(event, ActionExecuted):
state_features = tracker.feauturize_current_state(self.domain)
feature_vector = self.domain.slice_feature_history(
self.featurizer, state_features, self.config.max_history)
hashed = utils.HashableNDArray(feature_vector)
# only continue with trackers that created a
# featurization we haven't observed at this event
if (hashed not in featurizations
or not self.config.remove_duplicates):
featurizations.add(hashed)
if not event.unpredictable:
# only actions which can be predicted at a stories start
a_idx = self.domain.index_for_action(event.action_name)
features.append(feature_vector)
labels.append(a_idx)
unique_trackers.append(tracker)
else:
unique_trackers.append(tracker)
tracker.update(event)
if not isinstance(event, ActionExecuted):
action_name = tracker.previously_executed_action()
self.events_metadata[action_name].add(event)
return TrackerResult(features, labels, unique_trackers)
def _process_event_with_trackers(
self,
event, # type: Event
trackers, # type: List[FeaturizedTracker]
max_history # type: int
):
"""Logs an event to all trackers.
Removes trackers that create equal featurizations.
From multiple trackers that create equal featurizations
we only need to keep one. Because as we continue processing
events and story steps, all trackers that created the
same featurization once will do so in the future (as we
feed the same events to all trackers)."""
# collected trackers that created different featurizations
unique_trackers = []
featurizations = set()
# collected training data
training_features = []
training_labels = []
for tracker in trackers:
if isinstance(event, ActionExecuted):
state_features = tracker.feauturize_current_state(self.domain)
feature_vector = self.domain.slice_feature_history(
self.featurizer, state_features, max_history)
hashed = utils.HashableNDArray(feature_vector)
# only continue with trackers that created a
# featurization we haven't observed at this event
if hashed not in featurizations:
featurizations.add(hashed)
if not event.unpredictable:
# only actions which can be predicted at a stories start
training_features.append(feature_vector)
training_labels.append(
self.domain.index_for_action(event.action_name))
unique_trackers.append(tracker)
else:
unique_trackers.append(tracker)
tracker.update(event)
return training_features, training_labels, unique_trackers