def train_comparison_models(story_filename,
domain,
output_path=None,
exclusion_percentages=None,
policy_configs=None,
runs=None,
dump_stories=False,
kwargs=None):
"""Train multiple models for comparison of policies"""
for r in range(cmdline_args.runs):
logging.info("Starting run {}/{}".format(r + 1, cmdline_args.runs))
for i in exclusion_percentages:
current_round = cmdline_args.percentages.index(i) + 1
for policy_config in policy_configs:
policies = config.load(policy_config)
if len(policies) > 1:
raise ValueError("You can only specify one policy per "
"model for comparison")
policy_name = type(policies[0]).__name__
output = os.path.join(output_path, 'run_' + str(r + 1),
policy_name + str(current_round))
logging.info("Starting to train {} round {}/{}"
" with {}% exclusion".format(
policy_name, current_round,
len(exclusion_percentages), i))
train_dialogue_model(domain,
stories,
output,
policy_config=policy_config,
exclusion_percentage=i,
kwargs=kwargs,
dump_stories=dump_stories)
def visualize(config_path: Text, domain_path: Text, stories_path: Text,
nlu_data_path: Text, output_path: Text, max_history: int):
from rasa_core.agent import Agent
from rasa_core import config
policies = config.load(config_path)
agent = Agent(domain_path, policies=policies)
# this is optional, only needed if the `/greet` type of
# messages in the stories should be replaced with actual
# messages (e.g. `hello`)
if nlu_data_path is not None:
from rasa_nlu.training_data import load_data
nlu_data_path = load_data(nlu_data_path)
else:
nlu_data_path = None
logger.info("Starting to visualize stories...")
agent.visualize(stories_path,
output_path,
max_history,
nlu_training_data=nlu_data_path)
full_output_path = "file://{}".format(os.path.abspath(output_path))
logger.info(
"Finished graph creation. Saved into {}".format(full_output_path))
import webbrowser
webbrowser.open(full_output_path)
def train_dialogue_model(domain_file,
stories_file,
output_path,
interpreter=None,
endpoints=AvailableEndpoints(),
dump_stories=False,
policy_config=None,
exclusion_percentage=None,
kwargs=None):
if not kwargs:
kwargs = {}
policies = config.load(policy_config)
agent = Agent(domain_file,
generator=endpoints.nlg,
action_endpoint=endpoints.action,
interpreter=interpreter,
policies=policies)
data_load_args, kwargs = utils.extract_args(
kwargs, {
"use_story_concatenation", "unique_last_num_states",
"augmentation_factor", "remove_duplicates", "debug_plots"
})
training_data = agent.load_data(stories_file,
exclusion_percentage=exclusion_percentage,
**data_load_args)
agent.train(training_data, **kwargs)
agent.persist(output_path, dump_stories)
return agent
def train_dialogue_model(domain_file,
stories_file,
output_path,
interpreter=None,
endpoints=AvailableEndpoints(),
max_history=None,
dump_flattened_stories=False,
policy_config=None,
kwargs=None):
if not kwargs:
kwargs = {}
fallback_args, kwargs = utils.extract_args(
kwargs, {"nlu_threshold", "core_threshold", "fallback_action_name"})
policies = config.load(policy_config, fallback_args, max_history)
agent = Agent(domain_file,
generator=endpoints.nlg,
action_endpoint=endpoints.action,
interpreter=interpreter,
policies=policies)
data_load_args, kwargs = utils.extract_args(
kwargs, {
"use_story_concatenation", "unique_last_num_states",
"augmentation_factor", "remove_duplicates", "debug_plots"
})
training_data = agent.load_data(stories_file, **data_load_args)
agent.train(training_data, **kwargs)
agent.persist(output_path, dump_flattened_stories)
return agent
def train_core(domain_file, model_path, training_data_file, policy_config):
logging.basicConfig(filename=logfile, level=logging.DEBUG)
agent = Agent(domain_file, policies=config.load(policy_config))
training_data = agent.load_data(training_data_file)
agent.train(training_data)
agent.persist(model_path)
return agent
def train(domain_file: Text,
stories_file: Text,
output_path: Text,
interpreter: Optional[NaturalLanguageInterpreter] = None,
endpoints: AvailableEndpoints = AvailableEndpoints(),
dump_stories: bool = False,
policy_config: Text = None,
exclusion_percentage: int = None,
kwargs: Optional[Dict] = None):
from rasa_core.agent import Agent
if not kwargs:
kwargs = {}
policies = config.load(policy_config)
agent = Agent(domain_file,
generator=endpoints.nlg,
action_endpoint=endpoints.action,
interpreter=interpreter,
policies=policies)
data_load_args, kwargs = utils.extract_args(
kwargs, {
"use_story_concatenation", "unique_last_num_states",
"augmentation_factor", "remove_duplicates", "debug_plots"
})
training_data = agent.load_data(stories_file,
exclusion_percentage=exclusion_percentage,
**data_load_args)
agent.train(training_data, **kwargs)
agent.persist(output_path, dump_stories)
return agent
def run_weather_online(input_channel, interpreter,domain_file="weather_domain.yml",training_data_file='data/stories.md'):
policies2 = policy_config.load("config.yml")
agent = Agent("weather_domain.yml", policies=policies2)
data = asyncio.run(agent.load_data(training_data_file))
agent.train(data)
return agent
def run_weather_online(interpreter,
domain_file="weather_domain.yml",
training_data_file='data/stories.md'):
policies2 = policy_config.load("config.yml")
action_endpoint = "endpoint.yml"
agent = Agent(domain_file,policies=policies2,interpreter=interpreter,action_endpoint=action_endpoint)
data = asyncio.run(agent.load_data(training_data_file))
agent.train(data)
interactive.run_interactive_learning(agent,training_data_file)
return agent
def train_nlu():
from rasa_nlu.training_data import load_data
from rasa_nlu import config
from rasa_nlu.model import Trainer
training_data = load_data('data/rasa_dataset_training.json')
trainer = Trainer(config.load("configs/nlu_embedding_config.yml"))
trainer.train(training_data)
model_directory = trainer.persist('models/nlu/',
fixed_model_name="current")
return model_directory
def train_dialogue_transformer(domain_file="mobile_domain.yml",
model_path="models/dialogue_transformer",
training_data_file="data/mobile_edit_story.md"):
# 通过加载yml配置文件方式配置policy
policies = config.load('./policy/attention_policy.yml')
agent = Agent(domain_file,
policies=policies)
training_data = agent.load_data(training_data_file)
agent.train(
training_data,
validation_split=0.2
)
agent.persist(model_path)
return agent
def test_agent_and_persist():
policies = config.load("policies.yml")
policies[0] = KerasPolicy(epochs=2) # Keep training times low
agent = Agent("domain.yml", policies=policies)
training_data = agent.load_data("data/stories.md")
agent.train(training_data, validation_split=0.0)
agent.persist("./tests/models/dialogue")
loaded = Agent.load("./tests/models/dialogue")
assert agent.handle_text("/greet") is not None
assert loaded.domain.action_names == agent.domain.action_names
assert loaded.domain.intents == agent.domain.intents
assert loaded.domain.entities == agent.domain.entities
assert loaded.domain.templates == agent.domain.templates
utils.add_logging_option_arguments(parser)
cli.arguments.add_config_arg(parser, nargs=1)
cli.arguments.add_domain_arg(parser)
cli.arguments.add_model_and_story_group(parser,
allow_pretrained_model=False)
return parser
if __name__ == '__main__':
arg_parser = create_argument_parser()
cmdline_arguments = arg_parser.parse_args()
utils.configure_colored_logging(cmdline_arguments.loglevel)
policies = config.load(cmdline_arguments.config[0])
agent = Agent(cmdline_arguments.domain, policies=policies)
# this is optional, only needed if the `/greet` type of
# messages in the stories should be replaced with actual
# messages (e.g. `hello`)
if cmdline_arguments.nlu_data is not None:
from rasa_nlu.training_data import load_data
nlu_data = load_data(cmdline_arguments.nlu_data)
else:
nlu_data = None
stories = cli.stories_from_cli_args(cmdline_arguments)
import logging
from rasa_core import training
from rasa_core.actions import Action
from rasa_core.agent import Agent
from rasa_core.domain import Domain
from rasa_core.policies.keras_policy import KerasPolicy
from rasa_core.policies.memoization import MemoizationPolicy
from rasa_core.featurizers import MaxHistoryTrackerFeaturizer, BinarySingleStateFeaturizer
from rasa_core.interpreter import RegexInterpreter
from rasa_core.interpreter import RasaNLUInterpreter
from rasa_core.policies import FallbackPolicy, KerasPolicy, MemoizationPolicy, FormPolicy
fallback = FallbackPolicy(fallback_action_name="utter_default",
core_threshold=0.2,
nlu_threshold=0.1)
from rasa_core import config as policy_config
policies = policy_config.load("policy.yml")
# - name: KerasPolicy
# epochs: 100
# max_history: 3
# - name: MemoizationPolicy
# max_history: 3
# - name: FallbackPolicy
# nlu_threshold: 0.1
# core_threshold: 0.2
# fallback_action_name: 'utter_default'
# - name: FormPolicy
# Function
#------------
def test_load_config(filename):
loaded = load(filename, None, None)
assert len(loaded) == 2
assert isinstance(loaded[0], MemoizationPolicy)
assert isinstance(loaded[1], ExamplePolicy)
import logging
from rasa_core.agent import Agent
from rasa_core import config as policy_config
from rasa_core.policies.keras_policy import KerasPolicy
from rasa_core.policies.memoization import MemoizationPolicy
if __name__ == '__main__':
logging.basicConfig(level='INFO')
model_path = './models/dialogue'
policies = policy_config.load("./policies.yml")
agent = Agent('chat_domain.yml', policies=policies)
training_data_file = agent.load_data('./stories.md')
agent.train(training_data_file)
agent.persist(model_path)
def train_core(domain_file, model_path, training_data_file, policy_config):
agent = Agent(domain_file, policies=config.load(policy_config))
training_data = agent.load_data(training_data_file)
agent.train(training_data)
agent.persist(model_path)
return agent