Merge pull request #8141 from RasaHQ/improved-startup-time

Improved startup time

changelog/4280.improvement.md

@@ -0,0 +1 @@
+Improved CLI startup time.

changelog/8141.misc.md

@@ -0,0 +1,3 @@
+The following modules were renamed:
+* `rasa.train` -> `rasa.model_training`
+* `rasa.test` -> `rasa.model_testing`

changelog/8141.removal.md

@@ -0,0 +1,6 @@
+The following import abbreviations were removed:
+* `rasa.core.train`: Please use `rasa.core.train.train` instead.
+* `rasa.core.visualize`: Please use `rasa.core.visualize.visualize` instead.
+* `rasa.nlu.train`: Please use `rasa.nlu.train.train` instead.
+* `rasa.nlu.test`: Please use `rasa.nlu.test.run_evaluation` instead.
+* `rasa.nlu.cross_validate`: Please use `rasa.nlu.test.cross_validate` instead.

docs/docs/telemetry/telemetry.mdx

@@ -113,7 +113,6 @@ Here is an example report that shows the data reported to Rasa after running
     "project": "a0a7178e6e5f9e6484c5cfa3ea4497ffc0c96d0ad3f3ad8e9399a1edd88e3cf4",
     "python": "3.7.5",
     "rasa_open_source": "2.0.0",
-    "gpu": 0,
     "cpu": 16
   }
 }

rasa/__init__.py

@@ -1,12 +1,10 @@
 import logging
 
 from rasa import version
+from rasa.api import run, train, test
 
 # define the version before the other imports since these need it
 __version__ = version.__version__
 
-from rasa.run import run
-from rasa.train import train
-from rasa.test import test
 
 logging.getLogger(__name__).addHandler(logging.NullHandler())

rasa/api.py

@@ -0,0 +1,155 @@
+import rasa.shared.constants
+import typing
+
+# WARNING: Be careful about adding any top level imports at this place!
+#   These functions are imported in `rasa.__init__` and any top level import
+#   added here will get executed as soon as someone runs `import rasa`.
+#   Some imports are very slow (e.g. `tensorflow`) and we want them to get
+#   imported when running `import rasa`. If you add more imports here,
+#   please check that in the chain you are importing, no slow packages
+#   are getting imported.
+
+if typing.TYPE_CHECKING:
+    from typing import Any, Text, Dict, Union, List, Optional, NoReturn
+    from rasa.model_training import TrainingResult
+    import asyncio
+
+
+def run(
+    model: "Text",
+    endpoints: "Text",
+    connector: "Text" = None,
+    credentials: "Text" = None,
+    **kwargs: "Dict[Text, Any]",
+) -> "NoReturn":
+    """Runs a Rasa model.
+
+    Args:
+        model: Path to model archive.
+        endpoints: Path to endpoints file.
+        connector: Connector which should be use (overwrites `credentials`
+        field).
+        credentials: Path to channel credentials file.
+        **kwargs: Additional arguments which are passed to
+        `rasa.core.run.serve_application`.
+
+    """
+    import rasa.core.run
+    from rasa.core.utils import AvailableEndpoints
+    from rasa.shared.utils.cli import print_warning
+    import rasa.shared.utils.common
+    from rasa.shared.constants import DOCS_BASE_URL
+
+    _endpoints = AvailableEndpoints.read_endpoints(endpoints)
+
+    if not connector and not credentials:
+        connector = "rest"
+
+        print_warning(
+            f"No chat connector configured, falling back to the "
+            f"REST input channel. To connect your bot to another channel, "
+            f"read the docs here: {DOCS_BASE_URL}/messaging-and-voice-channels"
+        )
+
+    kwargs = rasa.shared.utils.common.minimal_kwargs(
+        kwargs, rasa.core.run.serve_application
+    )
+    rasa.core.run.serve_application(
+        model,
+        channel=connector,
+        credentials=credentials,
+        endpoints=_endpoints,
+        **kwargs,
+    )
+
+
+def train(
+    domain: "Text",
+    config: "Text",
+    training_files: "Union[Text, List[Text]]",
+    output: "Text" = rasa.shared.constants.DEFAULT_MODELS_PATH,
+    dry_run: bool = False,
+    force_training: bool = False,
+    fixed_model_name: "Optional[Text]" = None,
+    persist_nlu_training_data: bool = False,
+    core_additional_arguments: "Optional[Dict]" = None,
+    nlu_additional_arguments: "Optional[Dict]" = None,
+    loop: "Optional[asyncio.AbstractEventLoop]" = None,
+    model_to_finetune: "Optional[Text]" = None,
+    finetuning_epoch_fraction: float = 1.0,
+) -> "TrainingResult":
+    """Runs Rasa Core and NLU training in `async` loop.
+
+    Args:
+        domain: Path to the domain file.
+        config: Path to the config for Core and NLU.
+        training_files: Paths to the training data for Core and NLU.
+        output: Output path.
+        dry_run: If `True` then no training will be done, and the information about
+            whether the training needs to be done will be printed.
+        force_training: If `True` retrain model even if data has not changed.
+        fixed_model_name: Name of model to be stored.
+        persist_nlu_training_data: `True` if the NLU training data should be persisted
+            with the model.
+        core_additional_arguments: Additional training parameters for core training.
+        nlu_additional_arguments: Additional training parameters forwarded to training
+            method of each NLU component.
+        loop: Optional EventLoop for running coroutines.
+        model_to_finetune: Optional path to a model which should be finetuned or
+            a directory in case the latest trained model should be used.
+        finetuning_epoch_fraction: The fraction currently specified training epochs
+            in the model configuration which should be used for finetuning.
+
+    Returns:
+        An instance of `TrainingResult`.
+    """
+    from rasa.model_training import train_async
+    import rasa.utils.common
+
+    return rasa.utils.common.run_in_loop(
+        train_async(
+            domain=domain,
+            config=config,
+            training_files=training_files,
+            output=output,
+            dry_run=dry_run,
+            force_training=force_training,
+            fixed_model_name=fixed_model_name,
+            persist_nlu_training_data=persist_nlu_training_data,
+            core_additional_arguments=core_additional_arguments,
+            nlu_additional_arguments=nlu_additional_arguments,
+            model_to_finetune=model_to_finetune,
+            finetuning_epoch_fraction=finetuning_epoch_fraction,
+        ),
+        loop,
+    )
+
+
+def test(
+    model: "Text",
+    stories: "Text",
+    nlu_data: "Text",
+    output: "Text" = rasa.shared.constants.DEFAULT_RESULTS_PATH,
+    additional_arguments: "Optional[Dict]" = None,
+) -> None:
+    """Test a Rasa model against a set of test data.
+
+    Args:
+        model: model to test
+        stories: path to the dialogue test data
+        nlu_data: path to the NLU test data
+        output: path to folder where all output will be stored
+        additional_arguments: additional arguments for the test call
+    """
+    from rasa.model_testing import test_core
+    import rasa.utils.common
+    from rasa.model_testing import test_nlu
+
+    if additional_arguments is None:
+        additional_arguments = {}
+
+    test_core(model, stories, output, additional_arguments)
+
+    rasa.utils.common.run_in_loop(
+        test_nlu(model, nlu_data, output, additional_arguments)
+    )

rasa/cli/data.py

@@ -11,11 +11,6 @@
 from rasa.cli.arguments import data as arguments
 from rasa.cli.arguments import default_arguments
 import rasa.cli.utils
-from rasa.core.training.converters.responses_prefix_converter import (
-    DomainResponsePrefixConverter,
-    StoryResponsePrefixConverter,
-)
-import rasa.nlu.convert
 from rasa.shared.constants import (
     DEFAULT_DATA_PATH,
     DEFAULT_CONFIG_PATH,
@@ -24,6 +19,10 @@
 )
 import rasa.shared.data
 from rasa.shared.core.constants import (
+    POLICY_NAME_FALLBACK,
+    POLICY_NAME_FORM,
+    POLICY_NAME_MAPPING,
+    POLICY_NAME_TWO_STAGE_FALLBACK,
     USER_INTENT_OUT_OF_SCOPE,
     ACTION_DEFAULT_FALLBACK_NAME,
 )
@@ -38,18 +37,13 @@
 import rasa.shared.nlu.training_data.util
 import rasa.shared.utils.cli
 import rasa.utils.common
-from rasa.utils.converter import TrainingDataConverter
-from rasa.validator import Validator
 from rasa.shared.core.domain import Domain, InvalidDomain
 import rasa.shared.utils.io
-import rasa.core.config
-from rasa.core.policies.form_policy import FormPolicy
-from rasa.core.policies.fallback import FallbackPolicy
-from rasa.core.policies.two_stage_fallback import TwoStageFallbackPolicy
-from rasa.core.policies.mapping_policy import MappingPolicy
 
 if TYPE_CHECKING:
     from rasa.shared.core.training_data.structures import StoryStep
+    from rasa.validator import Validator
+    from rasa.utils.converter import TrainingDataConverter
 
 logger = logging.getLogger(__name__)
 
@@ -242,6 +236,7 @@ def validate_files(args: argparse.Namespace, stories_only: bool = False) -> None
         args: Commandline arguments
         stories_only: If `True`, only the story structure is validated.
     """
+    from rasa.validator import Validator
 
     config = rasa.cli.utils.get_validated_path(
         args.config, "config", DEFAULT_CONFIG_PATH, none_is_valid=True
@@ -278,15 +273,15 @@ def validate_stories(args: argparse.Namespace) -> None:
     validate_files(args, stories_only=True)
 
 
-def _validate_domain(validator: Validator) -> bool:
+def _validate_domain(validator: "Validator") -> bool:
     return validator.verify_domain_validity()
 
 
-def _validate_nlu(validator: Validator, args: argparse.Namespace) -> bool:
+def _validate_nlu(validator: "Validator", args: argparse.Namespace) -> bool:
     return validator.verify_nlu(not args.fail_on_warnings)
 
 
-def _validate_story_structure(validator: Validator, args: argparse.Namespace) -> bool:
+def _validate_story_structure(validator: "Validator", args: argparse.Namespace) -> bool:
     # Check if a valid setting for `max_history` was given
     if isinstance(args.max_history, int) and args.max_history < 1:
         raise argparse.ArgumentTypeError(
@@ -299,6 +294,8 @@ def _validate_story_structure(validator: Validator, args: argparse.Namespace) ->
 
 
 def _convert_nlu_data(args: argparse.Namespace) -> None:
+    import rasa.nlu.convert
+
     from rasa.nlu.training_data.converters.nlu_markdown_to_yaml_converter import (
         NLUMarkdownToYamlConverter,
     )
@@ -356,8 +353,14 @@ def _convert_nlg_data(args: argparse.Namespace) -> None:
 
 def _migrate_responses(args: argparse.Namespace) -> None:
     """Migrate retrieval intent responses to the new 2.0 format.
+
     It does so modifying the stories and domain files.
     """
+    from rasa.core.training.converters.responses_prefix_converter import (
+        DomainResponsePrefixConverter,
+        StoryResponsePrefixConverter,
+    )
+
     if args.format == "yaml":
         rasa.utils.common.run_in_loop(
             _convert_to_yaml(args.out, args.domain, DomainResponsePrefixConverter())
@@ -374,7 +377,7 @@ def _migrate_responses(args: argparse.Namespace) -> None:
 
 
 async def _convert_to_yaml(
-    out_path: Text, data_path: Text, converter: TrainingDataConverter
+    out_path: Text, data_path: Text, converter: "TrainingDataConverter"
 ) -> None:
 
     output = Path(out_path)
@@ -415,7 +418,7 @@ async def _convert_to_yaml(
 
 
 async def _convert_file_to_yaml(
-    source_file: Path, target_dir: Path, converter: TrainingDataConverter
+    source_file: Path, target_dir: Path, converter: "TrainingDataConverter"
 ) -> bool:
     """Converts a single training data file to `YAML` format.
 
@@ -447,6 +450,8 @@ def _migrate_model_config(args: argparse.Namespace) -> None:
     Args:
         args: The commandline args with the required paths.
     """
+    import rasa.core.config
+
     configuration_file = Path(args.config)
     model_configuration = _get_configuration(configuration_file)
 
@@ -499,32 +504,32 @@ def _get_configuration(path: Path) -> Dict:
     _assert_two_stage_fallback_policy_is_migratable(config)
     _assert_only_one_fallback_policy_present(policy_names)
 
-    if FormPolicy.__name__ in policy_names:
+    if POLICY_NAME_FORM in policy_names:
         _warn_about_manual_forms_migration()
 
     return config
 
 
 def _assert_config_needs_migration(policies: List[Text]) -> None:
     migratable_policies = {
-        MappingPolicy.__name__,
-        FallbackPolicy.__name__,
-        TwoStageFallbackPolicy.__name__,
+        POLICY_NAME_MAPPING,
+        POLICY_NAME_FALLBACK,
+        POLICY_NAME_TWO_STAGE_FALLBACK,
     }
 
     if not migratable_policies.intersection((set(policies))):
         rasa.shared.utils.cli.print_error_and_exit(
             f"No policies were found which need migration. This command can migrate "
-            f"'{MappingPolicy.__name__}', '{FallbackPolicy.__name__}' and "
-            f"'{TwoStageFallbackPolicy.__name__}'."
+            f"'{POLICY_NAME_MAPPING}', '{POLICY_NAME_FALLBACK}' and "
+            f"'{POLICY_NAME_TWO_STAGE_FALLBACK}'."
         )
 
 
 def _warn_about_manual_forms_migration() -> None:
     rasa.shared.utils.cli.print_warning(
-        f"Your model configuration contains the '{FormPolicy.__name__}'. "
-        f"Note that this command does not migrate the '{FormPolicy.__name__}' and "
-        f"you have to migrate the '{FormPolicy.__name__}' manually. "
+        f"Your model configuration contains the '{POLICY_NAME_FORM}'. "
+        f"Note that this command does not migrate the '{POLICY_NAME_FORM}' and "
+        f"you have to migrate the '{POLICY_NAME_FORM}' manually. "
         f"Please see the migration guide for further details: "
         f"{DOCS_URL_MIGRATION_GUIDE}"
     )
@@ -533,7 +538,7 @@ def _warn_about_manual_forms_migration() -> None:
 def _assert_nlu_pipeline_given(config: Dict, policy_names: List[Text]) -> None:
     if not config.get("pipeline") and any(
         policy in policy_names
-        for policy in [FallbackPolicy.__name__, TwoStageFallbackPolicy.__name__]
+        for policy in [POLICY_NAME_FALLBACK, POLICY_NAME_TWO_STAGE_FALLBACK]
     ):
         rasa.shared.utils.cli.print_error_and_exit(
             "The model configuration has to include an NLU pipeline. This is required "
@@ -546,7 +551,7 @@ def _assert_two_stage_fallback_policy_is_migratable(config: Dict) -> None:
         (
             policy_config
             for policy_config in config.get("policies", [])
-            if policy_config.get("name") == TwoStageFallbackPolicy.__name__
+            if policy_config.get("name") == POLICY_NAME_TWO_STAGE_FALLBACK
         ),
         None,
     )
@@ -583,10 +588,7 @@ def _assert_two_stage_fallback_policy_is_migratable(config: Dict) -> None:
 
 
 def _assert_only_one_fallback_policy_present(policies: List[Text]) -> None:
-    if (
-        FallbackPolicy.__name__ in policies
-        and TwoStageFallbackPolicy.__name__ in policies
-    ):
+    if POLICY_NAME_FALLBACK in policies and POLICY_NAME_TWO_STAGE_FALLBACK in policies:
         rasa.shared.utils.cli.print_error_and_exit(
             "Your policy configuration contains two configured policies for handling "
             "fallbacks. Please decide on one."