async def load_data(
resource_name: Union[Text, "TrainingDataImporter"],
domain: "Domain",
remove_duplicates: bool = True,
unique_last_num_states: Optional[int] = None,
augmentation_factor: int = 50,
tracker_limit: Optional[int] = None,
use_story_concatenation: bool = True,
debug_plots: bool = False,
exclusion_percentage: Optional[int] = None,
) -> List["DialogueStateTracker"]:
"""
Load training data from a resource.
Args:
resource_name: resource to load the data from. either a path or an importer
domain: domain used for loading
remove_duplicates: should duplicated training examples be removed?
unique_last_num_states: number of states in a conversation that make the
a tracker unique (this is used to identify duplicates)
augmentation_factor:
by how much should the story training data be augmented
tracker_limit:
maximum number of trackers to generate during augmentation
use_story_concatenation:
should stories be concatenated when doing data augmentation
debug_plots:
generate debug plots during loading
exclusion_percentage:
how much data to exclude
Returns:
list of loaded trackers
"""
from rasa.core.training.generator import TrainingDataGenerator
from rasa.importers.importer import TrainingDataImporter
if resource_name:
if isinstance(resource_name, TrainingDataImporter):
graph = await resource_name.get_stories(
exclusion_percentage=exclusion_percentage)
else:
graph = await extract_story_graph(
resource_name,
domain,
exclusion_percentage=exclusion_percentage)
g = TrainingDataGenerator(
graph,
domain,
remove_duplicates,
unique_last_num_states,
augmentation_factor,
tracker_limit,
use_story_concatenation,
debug_plots,
)
return g.generate()
else:
return []
async def load_data(
resource_name: Text,
domain: "Domain",
remove_duplicates: bool = True,
unique_last_num_states: Optional[int] = None,
augmentation_factor: int = 20,
tracker_limit: Optional[int] = None,
use_story_concatenation: bool = True,
debug_plots=False,
exclusion_percentage: int = None,
) -> List["DialogueStateTracker"]:
from rasa.core.training import extract_story_graph
from rasa.core.training.generator import TrainingDataGenerator
if resource_name:
graph = await extract_story_graph(
resource_name, domain, exclusion_percentage=exclusion_percentage)
g = TrainingDataGenerator(
graph,
domain,
remove_duplicates,
unique_last_num_states,
augmentation_factor,
tracker_limit,
use_story_concatenation,
debug_plots,
)
return g.generate()
else:
return []
async def visualize_stories(
story_steps: List[StoryStep],
domain: Domain,
output_file: Optional[Text],
max_history: int,
interpreter: NaturalLanguageInterpreter = RegexInterpreter(),
nlu_training_data: Optional["TrainingData"] = None,
should_merge_nodes: bool = True,
fontsize: int = 12,
silent: bool = False):
"""Given a set of stories, generates a graph visualizing the flows in the
stories.
Visualization is always a trade off between making the graph as small as
possible while
at the same time making sure the meaning doesn't change to "much". The
algorithm will
compress the graph generated from the stories to merge nodes that are
similar. Hence,
the algorithm might create paths through the graph that aren't actually
specified in the
stories, but we try to minimize that.
Output file defines if and where a file containing the plotted graph
should be stored.
The history defines how much 'memory' the graph has. This influences in
which situations the
algorithm will merge nodes. Nodes will only be merged if they are equal
within the history, this
means the larger the history is we take into account the less likely it
is we merge any nodes.
The training data parameter can be used to pass in a Rasa NLU training
data instance. It will
be used to replace the user messages from the story file with actual
messages from the training data."""
story_graph = StoryGraph(story_steps)
g = TrainingDataGenerator(story_graph,
domain,
use_story_concatenation=False,
tracker_limit=100,
augmentation_factor=0)
completed_trackers = g.generate(silent)
event_sequences = [t.events for t in completed_trackers]
graph = await visualize_neighborhood(None,
event_sequences,
output_file,
max_history,
interpreter,
nlu_training_data,
should_merge_nodes,
max_distance=1,
fontsize=fontsize)
return graph
async def train(self):
"""Train the engine.
"""
nltk.download('punkt')
lang = self.config['language']
if not os.path.exists('data/' + self.config['skill-id']):
_LOGGER.info("Starting Skill training.")
_LOGGER.info("Generating stories.")
data, domain_data, stories = await GenerateStories.run(
self.config['skill-id'], self.config['language'], self.asm)
training_data = TrainingData(training_examples=data)
nlu_config = RasaNLUModelConfig({
"language": lang,
"pipeline": self.config['pipeline'],
"data": None
})
trainer = Trainer(nlu_config, None, True)
_LOGGER.info("Training Arcus NLU")
trainer.train(training_data)
trainer.persist("data/" + self.config['skill-id'], None, 'nlu')
# Rasa core
domain = Domain.from_dict(domain_data)
reader = StoryFileReader(domain, RegexInterpreter(), None, False)
story_steps = await reader.process_lines(stories)
graph = StoryGraph(story_steps)
g = TrainingDataGenerator(
graph,
domain,
remove_duplicates=True,
unique_last_num_states=None,
augmentation_factor=20,
tracker_limit=None,
use_story_concatenation=True,
debug_plots=False,
)
training_trackers = g.generate()
policy_list = SimplePolicyEnsemble.from_dict(
{"policies": self.config['policies']})
policy_ensemble = SimplePolicyEnsemble(policy_list)
_LOGGER.info("Training Arcus Core")
policy_ensemble.train(training_trackers, domain)
policy_ensemble.persist(
"data/" + self.config['skill-id'] + "/core", False)
domain.persist("data/" + self.config['skill-id'] + "/core/model")
domain.persist_specification("data/" + self.config['skill-id'] +
"/core")
async def _generate_trackers(resource_name, agent,
max_stories=None,
use_e2e=False):
from rasa.core.training.generator import TrainingDataGenerator
from rasa.core import training
story_graph = await training.extract_story_graph(
resource_name, agent.domain, agent.interpreter, use_e2e)
g = TrainingDataGenerator(story_graph, agent.domain,
use_story_concatenation=False,
augmentation_factor=0,
tracker_limit=max_stories)
return g.generate()
def verify_story_structure(self,
ignore_warnings: bool = True,
max_history: Optional[int] = None) -> bool:
"""Verifies that the bot behaviour in stories is deterministic.
Args:
ignore_warnings: When `True`, return `True` even if conflicts were found.
max_history: Maximal number of events to take into account for conflict identification.
Returns:
`False` is a conflict was found and `ignore_warnings` is `False`.
`True` otherwise.
"""
logger.info("Story structure validation...")
trackers = TrainingDataGenerator(
self.story_graph,
domain=self.domain,
remove_duplicates=False,
augmentation_factor=0,
).generate()
# Create a list of `StoryConflict` objects
conflicts = rasa.core.training.story_conflict.find_story_conflicts(
trackers, self.domain, max_history)
if not conflicts:
logger.info("No story structure conflicts found.")
else:
for conflict in conflicts:
logger.warning(conflict)
return ignore_warnings or not conflicts
async def _generate_trackers(
resource_name: Text,
agent: "Agent",
max_stories: Optional[int] = None,
use_e2e: bool = False,
) -> List[Any]:
from rasa.core.training.generator import TrainingDataGenerator
from rasa.core import training
story_graph = await training.extract_story_graph(resource_name,
agent.domain, use_e2e)
g = TrainingDataGenerator(
story_graph,
agent.domain,
use_story_concatenation=False,
augmentation_factor=0,
tracker_limit=max_stories,
)
return g.generate_story_trackers()
async def _setup_trackers_for_testing(
domain_path: Text, training_data_file: Text
) -> Tuple[List[TrackerWithCachedStates], Domain]:
importer = RasaFileImporter(domain_path=domain_path,
training_data_paths=[training_data_file])
validator = await Validator.from_importer(importer)
trackers = TrainingDataGenerator(
validator.story_graph,
domain=validator.domain,
remove_duplicates=False,
augmentation_factor=0,
).generate()
return trackers, validator.domain
