improved build speed by reusing trained models

.github/workflows/continous-integration.yml

@@ -1,4 +1,4 @@
-name: Continous Integration
+name: Continuous Integration
 
 on: [push]
 
@@ -18,7 +18,7 @@ env:
 
 jobs:
   api:
-    name: API specification
+    name: Test API specification
     runs-on: ubuntu-latest
 
     steps:
@@ -177,6 +177,7 @@ jobs:
         GITHUB_TAG: ${{ github.ref }}
         GITHUB_REPO_SLUG: ${{ github.repository }}
       run: |
+        sudo apt-get update
         sudo apt-get -y install pandoc
         pip install -U github3.py pypandoc
         python3 scripts/publish_gh_release_notes.py

.github/workflows/documentation.yml

@@ -43,7 +43,7 @@ jobs:
         curl -sSL -o $RASABASTER "https://storage.googleapis.com/docs-theme/${RASABASTER}?q=$(date +%s%N)"
         pip install $RASABASTER
         pip install -r requirements.txt
-        pip install --no-cache-dir -r requirements-docs.txt
+        pip install -r requirements-docs.txt
         pip install git+https://${GITHUB_TOKEN}:x-oauth-basic@github.com/RasaHQ/sphinxcontrib-versioning.git@version_list
         pip install -e .
         pip list

docs/conf.py

@@ -336,7 +336,8 @@
 scv_whitelist_tags = (
     re.compile(r"^[2-9]+\.\d+\.\d+$"),
     re.compile(r"^1\.[1-9][0-9]+\.\d+$"),
-    re.compile(r"^1\.[6789]\.\d+$"),
+    re.compile(r"^1\.[789]\.\d+$"),
+    re.compile(r"^1\.6\.2$"),
     re.compile(r"^1\.5\.3$"),
     re.compile(r"^1\.4\.6$"),
     re.compile(r"^1\.3\.10$"),

scripts/publish_gh_release_notes.py

@@ -1,6 +1,6 @@
 """
 Script used to publish GitHub release notes extracted from CHANGELOG.rst.
-This script is executed by Github after a new release was successfully built.
+This script is executed by GitHub after a new release was successfully built.
 
 Uses the following environment variables:
 * GITHUB_TAG: the name of the tag of the current commit.

tests/cli/conftest.py

@@ -1,3 +1,8 @@
+import shutil
+
+from subprocess import check_call
+
+from _pytest.tmpdir import TempdirFactory
 from typing import Callable
 import pytest
 import os
@@ -22,10 +27,28 @@ def do_run(*args, stdin):
     return do_run
 
 
+@pytest.fixture(scope="session")
+def init_default_project(tmpdir_factory: TempdirFactory) -> str:
+    path = tmpdir_factory.mktemp("agent").strpath
+    os.environ["LOG_LEVEL"] = "ERROR"
+
+    check_call(["rasa", "init", "--no-prompt"], cwd=path)
+    return path
+
+
 @pytest.fixture
-def run_in_default_project(testdir: Testdir) -> Callable[..., RunResult]:
+def run_in_default_project(
+    testdir: Testdir, init_default_project: str
+) -> Callable[..., RunResult]:
     os.environ["LOG_LEVEL"] = "ERROR"
-    testdir.run("rasa", "init", "--no-prompt")
+
+    # makes sure we do not always retrain an initial model for every "new" project
+    for file_name in os.listdir(init_default_project):
+        full_file_name = os.path.join(init_default_project, file_name)
+        if os.path.isfile(full_file_name):
+            shutil.copy(full_file_name, str(testdir.tmpdir))
+        else:
+            shutil.copytree(full_file_name, str(testdir.tmpdir / file_name))
 
     def do_run(*args):
         args = ["rasa"] + list(args)

tests/conftest.py

@@ -1,9 +1,10 @@
-import logging
-from contextlib import contextmanager
-from typing import Text, List
+import asyncio
+from sanic.request import Request
+from sanic.testing import SanicTestClient
+
+from typing import Text, List, Tuple, Iterator
 
 import pytest
-from _pytest.logging import LogCaptureFixture
 from _pytest.tmpdir import TempdirFactory
 from sanic import Sanic
 
@@ -12,9 +13,11 @@
 from rasa.core.agent import Agent, load_agent
 from rasa.core.channels import channel
 from rasa.core.channels.channel import RestInput
+from rasa.core.domain import SessionConfig
 from rasa.core.policies import Policy
 from rasa.core.policies.memoization import AugmentedMemoizationPolicy
 from rasa.core.run import _create_app_without_api
+from rasa.core.tracker_store import InMemoryTrackerStore
 from rasa.model import get_model
 from rasa.train import train_async
 from rasa.utils.common import TempDirectoryPath
@@ -33,13 +36,22 @@
 # from a separatedly installable pytest-cli plugin.
 pytest_plugins = ["pytester"]
 
+# https://github.com/pytest-dev/pytest-asyncio/issues/68
+# this event_loop is used by pytest-asyncio, and redefining it
+# is currently the only way of changing the scope of this fixture
+@pytest.yield_fixture(scope="session")
+def event_loop(request: Request) -> Iterator[asyncio.AbstractEventLoop]:
+    loop = asyncio.get_event_loop_policy().new_event_loop()
+    yield loop
+    loop.close()
 
-@pytest.fixture
-async def default_agent(tmpdir_factory: TempdirFactory) -> Agent:
+
+@pytest.fixture(scope="session")
+async def _trained_default_agent(tmpdir_factory: TempdirFactory) -> Tuple[Agent, str]:
     model_path = tmpdir_factory.mktemp("model").strpath
 
     agent = Agent(
-        "data/test_domains/default.yml",
+        "data/test_domains/default_with_slots.yml",
         policies=[AugmentedMemoizationPolicy(max_history=3)],
     )
 
@@ -49,6 +61,18 @@ async def default_agent(tmpdir_factory: TempdirFactory) -> Agent:
     return agent
 
 
+def reset_conversation_state(agent: Agent) -> Agent:
+    # Clean tracker store after each test so tests don't affect each other
+    agent.tracker_store = InMemoryTrackerStore(agent.domain)
+    agent.domain.session_config = SessionConfig.default()
+    return agent
+
+
+@pytest.fixture
+async def default_agent(_trained_default_agent: Agent) -> Agent:
+    return reset_conversation_state(_trained_default_agent)
+
+
 @pytest.fixture(scope="session")
 async def trained_moodbot_path() -> Text:
     return await train_async(
@@ -66,17 +90,17 @@ async def unpacked_trained_moodbot_path(
     return get_model(trained_moodbot_path)
 
 
-@pytest.fixture
+@pytest.fixture(scope="session")
 async def stack_agent(trained_rasa_model: Text) -> Agent:
     return await load_agent(model_path=trained_rasa_model)
 
 
-@pytest.fixture
+@pytest.fixture(scope="session")
 async def core_agent(trained_core_model: Text) -> Agent:
     return await load_agent(model_path=trained_core_model)
 
 
-@pytest.fixture
+@pytest.fixture(scope="session")
 async def nlu_agent(trained_nlu_model: Text) -> Agent:
     return await load_agent(model_path=trained_nlu_model)
 
@@ -122,28 +146,26 @@ async def _train(*args, output_path=None, **kwargs):
     return _train
 
 
-@pytest.fixture()
+@pytest.fixture(scope="session")
 async def trained_rasa_model(
     trained_async,
     default_domain_path: Text,
-    default_config: List[Policy],
     default_nlu_data: Text,
     default_stories_file: Text,
 ) -> Text:
     trained_stack_model_path = await trained_async(
-        domain="data/test_domains/default.yml",
+        domain=default_domain_path,
         config=DEFAULT_STACK_CONFIG,
         training_files=[default_nlu_data, default_stories_file],
     )
 
     return trained_stack_model_path
 
 
-@pytest.fixture()
+@pytest.fixture(scope="session")
 async def trained_core_model(
     trained_async,
     default_domain_path: Text,
-    default_config: List[Policy],
     default_nlu_data: Text,
     default_stories_file: Text,
 ) -> Text:
@@ -156,7 +178,7 @@ async def trained_core_model(
     return trained_core_model_path
 
 
-@pytest.fixture()
+@pytest.fixture(scope="session")
 async def trained_nlu_model(
     trained_async,
     default_domain_path: Text,
@@ -208,7 +230,7 @@ async def rasa_server_without_api() -> Sanic:
     return app
 
 
-def get_test_client(server):
+def get_test_client(server: Sanic) -> SanicTestClient:
     test_client = server.test_client
     test_client.port = None
     return test_client

tests/core/conftest.py

@@ -1,9 +1,11 @@
 import asyncio
 import os
+
+from sanic.request import Request
 import uuid
 from datetime import datetime
 
-from typing import Text
+from typing import Text, Iterator
 
 import pytest
 from _pytest.tmpdir import TempdirFactory
@@ -96,7 +98,7 @@ def __init__(
 # this event_loop is used by pytest-asyncio, and redefining it
 # is currently the only way of changing the scope of this fixture
 @pytest.yield_fixture(scope="session")
-def event_loop(request):
+def event_loop(request: Request) -> Iterator[asyncio.AbstractEventLoop]:
     loop = asyncio.get_event_loop_policy().new_event_loop()
     yield loop
     loop.close()
@@ -136,56 +138,20 @@ def default_domain():
     return Domain.load(DEFAULT_DOMAIN_PATH_WITH_SLOTS)
 
 
-@pytest.fixture(scope="session")
-async def _default_agent(default_domain: Domain) -> Agent:
-    agent = Agent(
-        default_domain,
-        policies=[MemoizationPolicy()],
-        interpreter=RegexInterpreter(),
-        tracker_store=InMemoryTrackerStore(default_domain),
-    )
-    training_data = await agent.load_data(DEFAULT_STORIES_FILE)
-    agent.train(training_data)
-    return agent
-
-
-@pytest.fixture()
-async def default_agent(_default_agent: Agent) -> Agent:
-    # Clean tracker store after each test so tests don't affect each other
-    _default_agent.tracker_store = InMemoryTrackerStore(_default_agent.domain)
-    _default_agent.domain.session_config = SessionConfig.default()
-    return _default_agent
-
-
-@pytest.fixture(scope="session")
-def default_agent_path(_default_agent: Agent, tmpdir_factory: TempdirFactory):
-    path = tmpdir_factory.mktemp("agent").strpath
-    _default_agent.persist(path)
-    return path
-
-
 @pytest.fixture
 def default_channel() -> OutputChannel:
     return CollectingOutputChannel()
 
 
 @pytest.fixture
-async def default_processor(default_domain, default_nlg):
-    agent = Agent(
-        default_domain,
-        SimplePolicyEnsemble([AugmentedMemoizationPolicy()]),
-        interpreter=RegexInterpreter(),
-    )
-
-    training_data = await agent.load_data(DEFAULT_STORIES_FILE)
-    agent.train(training_data)
-    tracker_store = InMemoryTrackerStore(default_domain)
+async def default_processor(default_agent: Agent) -> MessageProcessor:
+    tracker_store = InMemoryTrackerStore(default_agent.domain)
     return MessageProcessor(
-        agent.interpreter,
-        agent.policy_ensemble,
-        default_domain,
+        default_agent.interpreter,
+        default_agent.policy_ensemble,
+        default_agent.domain,
         tracker_store,
-        default_nlg,
+        TemplatedNaturalLanguageGenerator(default_agent.domain.templates),
     )
 
 
@@ -243,39 +209,6 @@ def moodbot_metadata(unpacked_trained_moodbot_path):
     )
 
 
-@pytest.fixture()
-async def trained_stack_model(
-    trained_async,
-    default_domain_path,
-    default_stack_config,
-    default_nlu_data,
-    default_stories_file,
-):
-
-    trained_stack_model_path = await trained_async(
-        domain=default_domain_path,
-        config=default_stack_config,
-        training_files=[default_nlu_data, default_stories_file],
-    )
-
-    return trained_stack_model_path
-
-
-@pytest.fixture
-async def prepared_agent(tmpdir_factory) -> Agent:
-    model_path = tmpdir_factory.mktemp("model").strpath
-
-    agent = Agent(
-        "data/test_domains/default.yml",
-        policies=[AugmentedMemoizationPolicy(max_history=3)],
-    )
-
-    training_data = await agent.load_data(DEFAULT_STORIES_FILE)
-    agent.train(training_data)
-    agent.persist(model_path)
-    return agent
-
-
 @pytest.fixture
 def default_nlg(default_domain):
     return TemplatedNaturalLanguageGenerator(default_domain.templates)
@@ -297,30 +230,6 @@ def project() -> Text:
     return directory
 
 
-def train_model(loop, project: Text, filename: Text = "test.tar.gz"):
-    from rasa.constants import (
-        DEFAULT_CONFIG_PATH,
-        DEFAULT_DATA_PATH,
-        DEFAULT_DOMAIN_PATH,
-        DEFAULT_MODELS_PATH,
-    )
-    import rasa.train
-
-    output = os.path.join(project, DEFAULT_MODELS_PATH, filename)
-    domain = os.path.join(project, DEFAULT_DOMAIN_PATH)
-    config = os.path.join(project, DEFAULT_CONFIG_PATH)
-    training_files = os.path.join(project, DEFAULT_DATA_PATH)
-
-    rasa.train(domain, config, training_files, output, loop=loop)
-
-    return output
-
-
-@pytest.fixture(scope="session")
-def trained_model(loop, project) -> Text:
-    return train_model(loop, project)
-
-
 @pytest.fixture
 async def restaurantbot(trained_async) -> Text:
     restaurant_domain = os.path.join(RESTAURANTBOT_PATH, "domain.yml")