def test_trained_interpreter_passed_to_policies(tmp_path: Path,
monkeypatch: MonkeyPatch,
domain_path: Text,
stories_path: Text):
policies_config = {
"policies": [{
"name": TEDPolicy.__name__
}, {
"name": RulePolicy.__name__
}]
}
policy_train = Mock()
monkeypatch.setattr(TEDPolicy, "train", policy_train)
interpreter = Mock(spec=RasaNLUInterpreter)
train(
domain_path,
stories_path,
str(tmp_path),
interpreter=interpreter,
policy_config=policies_config,
additional_arguments={},
)
policy_train.assert_called_once()
assert policy_train.call_count == 1
_, _, kwargs = policy_train.mock_calls[0]
assert kwargs["interpreter"] == interpreter
def test_training_script_with_restart_stories(tmp_path: Path,
domain_path: Text):
train(
domain_path,
"data/test_yaml_stories/stories_restart.yml",
str(tmp_path),
interpreter=RegexInterpreter(),
policy_config="data/test_config/max_hist_config.yml",
additional_arguments={},
)
assert True
def test_training_script_with_max_history_set(tmp_path: Path,
domain_path: Text,
stories_path: Text):
tmpdir = str(tmp_path)
train(
domain_path,
stories_path,
tmpdir,
interpreter=RegexInterpreter(),
policy_config="data/test_config/max_hist_config.yml",
additional_arguments={},
)
agent = Agent.load(tmpdir)
expected_max_history = {RulePolicy: None}
for policy in agent.policy_ensemble.policies:
if hasattr(policy.featurizer, "max_history"):
expected_history = expected_max_history.get(type(policy), 5)
assert policy.featurizer.max_history == expected_history
def test_training_script_without_max_history_set(tmp_path: Path,
domain_path: Text,
stories_path: Text):
tmpdir = str(tmp_path)
train(
domain_path,
stories_path,
tmpdir,
interpreter=RegexInterpreter(),
policy_config="data/test_config/no_max_hist_config.yml",
additional_arguments={},
)
agent = Agent.load(tmpdir)
for policy in agent.policy_ensemble.policies:
if hasattr(policy.featurizer, "max_history"):
if type(policy) == MemoizationPolicy:
assert policy.featurizer.max_history == DEFAULT_MAX_HISTORY
else:
assert policy.featurizer.max_history is None
async def test_random_seed(tmp_path: Path, monkeypatch: MonkeyPatch,
domain_path: Text, stories_path: Text):
policies_config = {
"policies": [
{
"name": TEDPolicy.__name__,
"random_seed": 42
},
{
"name": RulePolicy.__name__
},
]
}
agent_1 = train(
domain_path,
stories_path,
str(tmp_path),
interpreter=RegexInterpreter(),
policy_config=policies_config,
additional_arguments={},
)
agent_2 = train(
domain_path,
stories_path,
str(tmp_path),
interpreter=RegexInterpreter(),
policy_config=policies_config,
additional_arguments={},
)
processor_1 = agent_1.create_processor()
processor_2 = agent_2.create_processor()
probs_1 = await processor_1.predict_next("1")
probs_2 = await processor_2.predict_next("2")
assert probs_1["confidence"] == probs_2["confidence"]