#901 - Dropped temp_primaite_sessiion_2 from conftest.py.

- Re-added the hard-coded mean rewards per episode values from a rpe-trained agent to the deterministic test in test_seeding_and_deterministic_session.py
- Partially tidies up some tests in test_observation_space.py; Still some work to be done on this at a later date.

src/primaite/config/_package_data/training/training_config_main.yaml

@@ -35,7 +35,7 @@ random_red_agent: False
 # Default is None (null)
 seed: null
 
-# Set whether the agent will be deterministic instead of stochastic
+# Set whether the agent evaluation will be deterministic instead of stochastic
 # Options are:
 # True
 # False

tests/config/ppo_not_seeded_training_config.yaml

@@ -35,7 +35,7 @@ random_red_agent: False
 # Default is None (null)
 seed: None
 
-# Set whether the agent will be deterministic instead of stochastic
+# Set whether the agent evaluation will be deterministic instead of stochastic
 # Options are:
 # True
 # False

tests/config/ppo_seeded_training_config.yaml

@@ -35,7 +35,7 @@ random_red_agent: False
 # Default is None (null)
 seed: 67890
 
-# Set whether the agent will be deterministic instead of stochastic
+# Set whether the agent evaluation will be deterministic instead of stochastic
 # Options are:
 # True
 # False
@@ -66,7 +66,7 @@ num_train_episodes: 10
 num_train_steps: 256
 
 # Number of episodes to run per session
-num_eval_episodes: 1
+num_eval_episodes: 5
 
 # Number of time_steps per episode
 num_eval_steps: 256

tests/conftest.py

@@ -119,60 +119,6 @@ def temp_primaite_session(request):
         return TempPrimaiteSession(training_config_path, lay_down_config_path)
 
 
-@pytest.fixture
-def temp_primaite_session_2(request):
-    """
-    Provides a temporary PrimaiteSession instance.
-
-    It's temporary as it uses a temporary directory as the session path.
-
-    To use this fixture you need to:
-
-    - parametrize your test function with:
-
-      - "temp_primaite_session"
-      - [[path to training config, path to lay down config]]
-    - Include the temp_primaite_session fixture as a param in your test
-    function.
-    - use the temp_primaite_session as a context manager assigning is the
-    name 'session'.
-
-    .. code:: python
-
-        from primaite.config.lay_down_config import dos_very_basic_config_path
-        from primaite.config.training_config import main_training_config_path
-        @pytest.mark.parametrize(
-            "temp_primaite_session",
-            [
-                [main_training_config_path(), dos_very_basic_config_path()]
-            ],
-            indirect=True
-        )
-        def test_primaite_session(temp_primaite_session):
-            with temp_primaite_session as session:
-                # Learning outputs are saved in session.learning_path
-                session.learn()
-
-                # Evaluation outputs are saved in session.evaluation_path
-                session.evaluate()
-
-                # To ensure that all files are written, you must call .close()
-                session.close()
-
-                # If you need to inspect any session outputs, it must be done
-                # inside the context manager
-
-            # Now that we've exited the context manager, the
-            # session.session_path directory and its contents are deleted
-    """
-    training_config_path = request.param[0]
-    lay_down_config_path = request.param[1]
-    with patch("primaite.agents.agent.get_session_path", get_temp_session_path) as mck:
-        mck.session_timestamp = datetime.now()
-
-        return TempPrimaiteSession(training_config_path, lay_down_config_path)
-
-
 @pytest.fixture
 def temp_session_path() -> Path:
     """

tests/test_observation_space.py

@@ -4,41 +4,9 @@ import numpy as np
 import pytest
 
 from primaite.environment.observations import NodeLinkTable, NodeStatuses, ObservationsHandler
-from primaite.environment.primaite_env import Primaite
 from tests import TEST_CONFIG_ROOT
 
 
-def run_generic_set_actions(env: Primaite):
-    """Run against a generic agent with specified blue agent actions."""
-    # Reset the environment at the start of the episode
-    # env.reset()
-    training_config = env.training_config
-    for episode in range(0, training_config.num_train_episodes):
-        for step in range(0, training_config.num_train_steps):
-            # Send the observation space to the agent to get an action
-            # TEMP - random action for now
-            # action = env.blue_agent_action(obs)
-            action = 0
-            print("Episode:", episode, "\nStep:", step)
-            if step == 2:
-                # [1, 1, 2, 1, 1, 1, 1(position)]
-                # NEED [1, 1, 1, 2, 1, 1, 1]
-                # Creates an ACL rule
-                # Allows traffic from server_1 to node_1 on port FTP
-                action = 43
-            elif step == 4:
-                action = 96
-
-            # Run the simulation step on the live environment
-            obs, reward, done, info = env.step(action)
-
-            # Break if done is True
-            if done:
-                break
-
-    return env
-
-
 @pytest.mark.parametrize(
     "temp_primaite_session",
     [
@@ -317,13 +285,9 @@ class TestAccessControlList:
         assert np.array_equal(obs, [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2])
 
     def test_observation_space_with_implicit_rule(self, temp_primaite_session):
-        """Test observation space is what is expected when an agent adds ACLs during an episode."""
-        # Used to use env from test fixture but AtrributeError function object has no 'training_config'
-        with temp_primaite_session as session:
-            env = session.env
-            env = run_generic_set_actions(env)
-            obs = env.env_obs
         """
+        Test observation space is what is expected when an agent adds ACLs during an episode.
+
         Observation space at the end of the episode.
         At the start of the episode, there is a single implicit Deny rule = 1,1,1,1,1,0
         (0 represents its initial position at top of ACL list)
@@ -333,6 +297,38 @@ class TestAccessControlList:
         On Step 7, there is a second rule added at POSITION 1: 2,4,2,3,3,1
         THINK THE RULES SHOULD BE THE OTHER WAY AROUND IN THE CURRENT OBSERVATION
         """
+        # TODO: Refactor this at some point to build a custom ACL Hardcoded
+        #  Agent and then patch the AgentIdentifier Enum class so that it
+        #  has ACL_AGENT. This then allows us to set the agent identified in
+        #  the main config and is a bit cleaner.
+        # Used to use env from test fixture but AtrributeError function object has no 'training_config'
+        with temp_primaite_session as session:
+            env = session.env
+
+            training_config = env.training_config
+            for episode in range(0, training_config.num_train_episodes):
+                for step in range(0, training_config.num_train_steps):
+                    # Send the observation space to the agent to get an action
+                    # TEMP - random action for now
+                    # action = env.blue_agent_action(obs)
+                    action = 0
+                    print("Episode:", episode, "\nStep:", step)
+                    if step == 2:
+                        # [1, 1, 2, 1, 1, 1, 1(position)]
+                        # NEED [1, 1, 1, 2, 1, 1, 1]
+                        # Creates an ACL rule
+                        # Allows traffic from server_1 to node_1 on port FTP
+                        action = 43
+                    elif step == 4:
+                        action = 96
+
+                    # Run the simulation step on the live environment
+                    obs, reward, done, info = env.step(action)
+
+                    # Break if done is True
+                    if done:
+                        break
+            obs = env.env_obs
         print("what i am testing", obs)
         # acl rule 1
         # source is 1 should be 4

tests/test_seeding_and_deterministic_session.py

@@ -1,5 +1,3 @@
-import time
-
 import pytest as pytest
 
 from primaite.config.lay_down_config import dos_very_basic_config_path
@@ -11,45 +9,45 @@ from tests import TEST_CONFIG_ROOT
     [[TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml", dos_very_basic_config_path()]],
     indirect=True,
 )
-@pytest.mark.parametrize(
+def test_seeded_learning(temp_primaite_session):
-    "temp_primaite_session_2",
+    """
-    [[TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml", dos_very_basic_config_path()]],
+    Test running seeded learning produces the same output when ran twice.
-    indirect=True,
+
-)
+    .. note::
-def test_seeded_learning(temp_primaite_session, temp_primaite_session_2):
+
-    """Test running seeded learning produces the same output when ran twice."""
+        If this is failing, the hard-coded expected_mean_reward_per_episode
+        from a pre-trained agent will probably need to be updated. If the
+        env changes and those changed how this agent is trained, chances are
+        the mean rewards are going to be different.
+
+        Run the test, but print out the session.learn_av_reward_per_episode()
+        before comparing it. Then copy the printed dict and replace the
+        expected_mean_reward_per_episode with those values. The test should
+        now work. If not, then you've got a bug :).
     """
     expected_mean_reward_per_episode = {
-        1: -90.703125,
+        1: -33.90625,
-        2: -91.15234375,
+        2: -32.32421875,
-        3: -87.5,
+        3: -25.234375,
-        4: -92.2265625,
+        4: -30.15625,
-        5: -94.6875,
+        5: -27.1484375,
-        6: -91.19140625,
+        6: -29.609375,
-        7: -88.984375,
+        7: -29.921875,
-        8: -88.3203125,
+        8: -29.3359375,
-        9: -112.79296875,
+        9: -28.046875,
-        10: -100.01953125,
+        10: -27.24609375,
     }
-    """
+
     with temp_primaite_session as session:
-        assert session._training_config.seed == 67890, (
+        assert (
-            "Expected output is based upon a agent that was trained with " "seed 67890"
+            session._training_config.seed == 67890
-        )
+        ), "Expected output is based upon a agent that was trained with seed 67890"
         session.learn()
-        actual_mean_reward_per_episode_run_1 = session.learn_av_reward_per_episode()
-    time.sleep(2)
-    with temp_primaite_session_2 as session:
-        assert session._training_config.seed == 67890, (
-            "Expected output is based upon a agent that was trained with " "seed 67890"
-        )
-        session.learn()
-        actual_mean_reward_per_episode_run_2 = session.learn_av_reward_per_episode()
 
-    assert actual_mean_reward_per_episode_run_1 == actual_mean_reward_per_episode_run_2
+        assert expected_mean_reward_per_episode == session.learn_av_reward_per_episode()
 
 
-@pytest.mark.skip(reason="Inconsistent results. Needs someone with RL " "knowledge to investigate further.")
+@pytest.mark.skip(reason="Inconsistent results. Needs someone with RL knowledge to investigate further.")
 @pytest.mark.parametrize(
     "temp_primaite_session",
     [[TEST_CONFIG_ROOT / "ppo_seeded_training_config.yaml", dos_very_basic_config_path()]],

tests/test_single_action_space.py

@@ -58,6 +58,10 @@ def run_generic_set_actions(env: Primaite):
 )
 def test_single_action_space_is_valid(temp_primaite_session):
     """Test single action space is valid."""
+    # TODO: Refactor this at some point to build a custom ACL Hardcoded
+    #  Agent and then patch the AgentIdentifier Enum class so that it
+    #  has ACL_AGENT. This then allows us to set the agent identified in
+    #  the main config and is a bit cleaner.
     with temp_primaite_session as session:
         env = session.env
 
@@ -95,6 +99,10 @@ def test_single_action_space_is_valid(temp_primaite_session):
 )
 def test_agent_is_executing_actions_from_both_spaces(temp_primaite_session):
     """Test to ensure the blue agent is carrying out both kinds of operations (NODE & ACL)."""
+    # TODO: Refactor this at some point to build a custom ACL Hardcoded
+    #  Agent and then patch the AgentIdentifier Enum class so that it
+    #  has ACL_AGENT. This then allows us to set the agent identified in
+    #  the main config and is a bit cleaner.
     with temp_primaite_session as session:
         env = session.env
         # Run environment with specified fixed blue agent actions only