tests

Limmen · Feb 4, 2024 · 900bf2b · 900bf2b
1 parent 8abe241
commit 900bf2b
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 6 deletions.
diff --git a/examples/eval/cyborg_scenario_two/evaluate_on_base_env.py b/examples/eval/cyborg_scenario_two/evaluate_on_base_env.py
@@ -7,7 +7,7 @@
 from gym_csle_cyborg.envs.cyborg_scenario_two_defender import CyborgScenarioTwoDefender
 
 if __name__ == '__main__':
-    ppo_policy = MetastoreFacade.get_ppo_policy(id=58)
+    ppo_policy = MetastoreFacade.get_ppo_policy(id=3)
     config = CSLECyborgConfig(
         gym_env_name="csle-cyborg-scenario-two-v1", scenario=2, baseline_red_agents=[RedAgentType.B_LINE_AGENT],
         maximum_steps=100, red_agent_distribution=[1.0], reduced_action_space=True, decoy_state=True,

diff --git a/examples/plotting/backup_orlando.py b/examples/plotting/backup_orlando.py
@@ -14,6 +14,7 @@
     # experiment_apple_bline_256.to_json_file("/home/kim/orlando_results/15_jan/apple_b_line_256.json")
     # experiment_apple_meander_256.to_json_file("/home/kim/orlando_results/15_jan/apple_meander_256.json")
     # experiment_cardiff_b_line_64.to_json_file("/home/kim/orlando_results/16_jan/cardiff_b_line_64.json")
+    experiment_cardiff_b_line_64.to_json_file("/home/kim/orlando_results/16_jan/cardiff_b_line_64_2.json")
     # experiment_cardiff_meander_64.to_json_file("/home/kim/orlando_results/16_jan/cardiff_meander_64.json")
 
     metric = "average_return"
@@ -41,6 +42,7 @@
     # print(avg_returns_means.shape)
     # avg_returns_stds = np.array(list(map(lambda x: PlottingUtil.mean_confidence_interval(
     #     data=x, confidence=confidence)[1], returns)))
+    b_line_opt = -13.73
     for i in range(len(avg_returns_means)):
         print(f"{i + 1} {avg_returns_means[i]} {avg_returns_means[i] + avg_returns_stds[i]} "
               f"{avg_returns_means[i] - avg_returns_stds[i]}")
diff --git a/examples/training/pomcp/cyborg_scenario_two_defender/run_cyborg_version_two_b_line.py b/examples/training/pomcp/cyborg_scenario_two_defender/run_cyborg_version_two_b_line.py
@@ -129,6 +129,6 @@
         player_type=PlayerType.DEFENDER, player_idx=0
     )
     agent = POMCPAgent(emulation_env_config=emulation_env_config, simulation_env_config=simulation_env_config,
-                       experiment_config=experiment_config, save_to_metastore=True)
+                       experiment_config=experiment_config, save_to_metastore=False)
     experiment_execution = agent.train()
     MetastoreFacade.save_experiment_execution(experiment_execution)
diff --git a/simulation-system/libs/csle-agents/src/csle_agents/agents/pomcp/pomcp.py b/simulation-system/libs/csle-agents/src/csle_agents/agents/pomcp/pomcp.py
@@ -360,8 +360,9 @@ def update_tree_with_new_samples(self, action_sequence: List[int], observation:
                     count = 0
                 else:
                     count += 1
-                if count >= 20000:
-                    raise ValueError(f"Invalid observation: {o} given state: {root.sample_state()}")
+                if count >= 80000:
+                    raise ValueError(f"Invalid observation: {o} given state: {root.sample_state()}, "
+                                     f"{root.sample_state()}, {root.sample_state()}")
             new_root.particles += particles
 
         # We now prune the old root from the tree

diff --git a/simulation-system/libs/csle-agents/src/csle_agents/agents/pomcp/pomcp_agent.py b/simulation-system/libs/csle-agents/src/csle_agents/agents/pomcp/pomcp_agent.py
@@ -275,11 +275,14 @@ def pomcp(self, exp_result: ExperimentResult, seed: int,
                 action_sequence.append(action)
                 s_prime = info[agents_constants.COMMON.STATE]
                 obs_id = info[agents_constants.COMMON.OBSERVATION]
+                print(eval_env.get_true_table())
+                print(eval_env.get_table())
+                from gym_csle_cyborg.util.cyborg_env_util import CyborgEnvUtil
+                print(CyborgEnvUtil.state_id_to_state_vector(state_id=s_prime))
+                print(CyborgEnvUtil.state_id_to_state_vector(state_id=obs_id, observation=True))
                 pomcp.update_tree_with_new_samples(action_sequence=action_sequence, observation=obs_id)
                 R += r
                 t += 1
-                print(eval_env.get_true_table())
-                print(eval_env.get_table())
                 if t % log_steps_frequency == 0:
                     rollout_action = -1
                     if rollout_policy is not None: