gr-libs 0.1.5__py3-none-any.whl → 0.1.6.post1__py3-none-any.whl

gr_libs/recognizer/recognizer.py

@@ -1,6 +1,5 @@
 from abc import ABC, abstractmethod
 from typing import List, Type
-
 from gr_libs.environment.environment import EnvProperty, SUPPORTED_DOMAINS
 from gr_libs.environment.utils.utils import domain_to_env_property
 from gr_libs.ml.base.rl_agent import RLAgent

{gr_libs-0.1.5.dist-info → gr_libs-0.1.6.post1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gr_libs
-Version: 0.1.5
+Version: 0.1.6.post1
 Summary: Package with goal recognition frameworks baselines
 Author: Ben Nageris
 Author-email: Matan Shamir <matan.shamir@live.biu.ac.il>, Osher Elhadad <osher.elhadad@live.biu.ac.il>
@@ -17,6 +17,7 @@ Requires-Dist: torchvision
 Requires-Dist: rl_zoo3
 Requires-Dist: stable_baselines3[extra]
 Requires-Dist: sb3_contrib
+Requires-Dist: pytest
 Provides-Extra: minigrid
 Requires-Dist: gr_envs[minigrid]; extra == "minigrid"
 Provides-Extra: highway
@@ -111,6 +112,25 @@ After installing GRLib, you will have access to custom Gym environments, allowin
 
 Tutorials demonstrating basic ODGR scenarios is available in the sub-package `tutorials`. These tutorials walk through the initialization and deployment process, showcasing how different GR algorithms adapt to emerging goals in various Gym environments.
 
+## Working with an initial dataset of trained agents
+gr_libs also includes a library of trained agents for the various supported environments within the package.
+To get the dataset of trained agents, you can run:
+```sh
+python download_dataset.py
+```
+
+An alternative is to use our docker image, which includes the dataset in it.
+You can:
+1. pull the image:
+```sh
+docker pull ghcr.io/MatanShamir1/gr_test_base:latest
+```
+2. run a container:
+```sh
+docker run -it ghcr.io/MatanShamir1/gr_test_base:latest bash
+```
+3. don't forget to install the package from within the container, go back to 'Setup' for that.
+
 ### Method 1: Writing a Custom Script
 
 1. **Create a recognizer**
@@ -118,6 +138,7 @@ Tutorials demonstrating basic ODGR scenarios is available in the sub-package `tu
    Specify the domain name and specific environment for the recognizer, effectively telling it the domain theory - the collection of states and actions in the environment.
 
    ```python
+   import gr_libs.environment # Triggers gym env registration - you must run it!
    recognizer = Graql(
        domain_name="minigrid",
        env_name="MiniGrid-SimpleCrossingS13N4"

{gr_libs-0.1.5.dist-info → gr_libs-0.1.6.post1.dist-info}/RECORD

@@ -1,3 +1,5 @@
+CI/README.md,sha256=CbWNAWrXFFwYq3sWAORhoQIE5busoNyYh_rFWVH1enw,800
+CI/docker_build_context/Dockerfile,sha256=Rk7LYTxOW7VVJcmNa8csZ4BwkunMYIiHX4WVSuMam50,311
 evaluation/analyze_results_cross_alg_cross_domain.py,sha256=s_DDh4rNfRnvQ0PDa2d5411jYOa7CaI1YeB8Dpup7QU,9803
 evaluation/create_minigrid_map_image.py,sha256=jaSW3n3tY222iFUeAMqedBP9cvD88GCzPrQ6_XHv5oQ,1242
 evaluation/file_system.py,sha256=SSYnj8QGFkq-8V_0s7x2MWbD88aFaoFY4Ogc_Pt8m6U,1601
@@ -8,12 +10,13 @@ evaluation/generate_task_specific_statistics_plots.py,sha256=rBsqaMe2irP_Cfo-icw
 evaluation/get_plans_images.py,sha256=BT-bGWuOPUAYpZVDwk7YMRBLdgKaDbNOBjMrtcl1Vjk,2346
 evaluation/increasing_and_decreasing_.py,sha256=fu1hkEjhOQC3jEsjiS7emW_UPRpVFCaae0d0E2MGZqI,2991
 gr_libs/__init__.py,sha256=-uKsQiHIL7yojbDwlTR-I8sj1WX9XT52PoFbPjtUTKo,145
-gr_libs/environment/__init__.py,sha256=oxEKmdvzQLKbbMxedqEf3bGsSJvp2XL9Bxr5JEO6a5o,1038
-gr_libs/environment/environment.py,sha256=Ca9m6W8KEt4le0HFSAUvSHW5lSHg_wwJaBqFSq1KlUg,6781
+gr_libs/_version.py,sha256=C8Me-BH17Mqlv65Ba3Tqc5gFEzabp8fxxyIA9C_XdDQ,517
+gr_libs/environment/__init__.py,sha256=HFVGBcufWf8-ahCo6h_s2pFEyvDy59cFg8z908RgdYo,1038
+gr_libs/environment/environment.py,sha256=d6ZbiAQ4H1aLrUFI8sm0BN9DVW3JtzpkodSi_70Z_PY,6780
 gr_libs/environment/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 gr_libs/environment/utils/utils.py,sha256=4yM3s30KjyuEmWR8UuICE5rR03zsLi3tzqNDvBkdPcU,537
 gr_libs/metrics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-gr_libs/metrics/metrics.py,sha256=IYPL90tuxbTRaNLaFvgzd5SMUwS8gsSANuWZcSVuKkM,8737
+gr_libs/metrics/metrics.py,sha256=4bnvs5suv-QrK9i1NuOzkE_E8uIzS1nlEazNDRXvZGs,8700
 gr_libs/ml/__init__.py,sha256=jrjxYqvSRgWwFWw7XQP9DzOwvmprMZ2umwT7t-DYtDU,233
 gr_libs/ml/agent.py,sha256=DSnK8nRx9SS76fAOZZEEvA68_meLjzm9lfQpMUXmGQU,1957
 gr_libs/ml/consts.py,sha256=mrbZk8n6QoGzLGaKmaxq4QlAsBbk4fhkCgXLuO9jXKw,365
@@ -42,8 +45,10 @@ gr_libs/ml/utils/format.py,sha256=nu7RzVwn_raG_fqqmnqlJgUjtA0yzKztkB3a5QZnRYo,30
 gr_libs/ml/utils/math.py,sha256=n62zssVOLHnUb4dPofAoFhoLOKl5n_xBzaKQOUQBoNc,440
 gr_libs/ml/utils/other.py,sha256=HKUfeLBbd4DgJxSTs3ya9KQ85Acx4TjycRrtGD9WQ3s,505
 gr_libs/ml/utils/storage.py,sha256=oCdvL_ypCglnSJsyyXzNyV_UJASTfioa3yJhFlFso64,4277
+gr_libs/problems/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+gr_libs/problems/consts.py,sha256=ON7yfKTAKETg7i3okDYuOzEU7KWvynyubl0m7TlU6Hs,38808
 gr_libs/recognizer/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-gr_libs/recognizer/recognizer.py,sha256=2lMlJNE7X13Y3FMpNfql-WAYw23NImIiomnVmCI75bM,1706
+gr_libs/recognizer/recognizer.py,sha256=ZrApJVdBQxKRYhhDiWLCNGmlxgi674nwgb30BgVggC8,1705
 gr_libs/recognizer/recognizer_doc.md,sha256=RnTvbZhl2opvU7-QT4pULCV5HCdJTw2dsu8WQOOiR3E,2521
 gr_libs/recognizer/gr_as_rl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 gr_libs/recognizer/gr_as_rl/gr_as_rl_recognizer.py,sha256=84GdfohC2dZoNH_QEo7GpSt8nZWdfqSRKCTY99X_iME,5215
@@ -52,12 +57,14 @@ gr_libs/recognizer/graml/gr_dataset.py,sha256=lG6m3ulxFELpH1oURnlcmNDWOrxyuzvlAR
 gr_libs/recognizer/graml/graml_recognizer.py,sha256=SGs7rtkA73lbCv9HISa6dfjVUJUhlH54QriVsoGVRss,15672
 gr_libs/recognizer/utils/__init__.py,sha256=ewSroxL7aATvvm-Xzc1_-61mP2LU2U28YaOEqvVVDB0,41
 gr_libs/recognizer/utils/format.py,sha256=e0AnqtPeYoJsV9Z7cEBpgbzTM0hLNxFIjn07fQ3YbQw,492
-tutorials/graml_minigrid_tutorial.py,sha256=0jSlsKd0H3DXA7rPSnw09y56pTSXvtXCFOKSuUvfDjs,1597
-tutorials/graml_panda_tutorial.py,sha256=LwNQPb7Kdg7X8jY7Zk13-8uBfDP8LMNzwgH-u3KOcjw,1861
-tutorials/graml_parking_tutorial.py,sha256=fsLbASIESUGnZe09eEhMcqxpU0NP8k1IQyGgJq_AFVs,1549
-tutorials/graml_point_maze_tutorial.py,sha256=gY8GCHnq32xyY7gSw3i3DL98TlfwgMkhO17csyV2QBA,1631
-tutorials/graql_minigrid_tutorial.py,sha256=VoXbEgL_hjQLfau6WohXxPK8rrv1VLA874F8PZ7ZtPk,1421
-gr_libs-0.1.5.dist-info/METADATA,sha256=h4QUMjuxouD3o1iKg2F3doJbKLbezkV1FaGKm1oBL0o,8905
-gr_libs-0.1.5.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-gr_libs-0.1.5.dist-info/top_level.txt,sha256=k7_l789QSJEr9JrtvsRMxNoTIDwNduq8mhIN-YoPJUM,29
-gr_libs-0.1.5.dist-info/RECORD,,
+tests/test_graml.py,sha256=ZJB2jqtf4Q2-KZredkJq90teqmHBIvigCAQpvR5G110,559
+tests/test_graql.py,sha256=-onMi13e2wStOmB5bYv2f3Ita3QFFiw416XMBkby0OI,141
+tutorials/graml_minigrid_tutorial.py,sha256=ONvxFi79R7d8dcd6gy083Z_yy9A2flhGTDIDRxurdx8,1782
+tutorials/graml_panda_tutorial.py,sha256=wtv_lsw0vsU7j45GKeWecTfE7jzfh4iVGEVnQyaWthM,2063
+tutorials/graml_parking_tutorial.py,sha256=46-sfxmYA9jLRSpqIF9z69MLSfOSTJarfjlQ_Igq294,1769
+tutorials/graml_point_maze_tutorial.py,sha256=mYq3IxYbf9jidq-4VdT3MdStV80Q5lytFv6Xzzn22Ys,1835
+tutorials/graql_minigrid_tutorial.py,sha256=Jb0TCUhiZQkFeafJWUTPnCISd4FKfPrqP-xfHiqCGKE,1635
+gr_libs-0.1.6.post1.dist-info/METADATA,sha256=UPwlwVlbGTpTsUhYwWH5hYr-hSBpcWjrFIA7sWg0Kj4,9620
+gr_libs-0.1.6.post1.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
+gr_libs-0.1.6.post1.dist-info/top_level.txt,sha256=rL-bbK-KnLzVbLIUCdN1riH58lup3jG0NJ3LTt_qSwo,38
+gr_libs-0.1.6.post1.dist-info/RECORD,,

{gr_libs-0.1.5.dist-info → gr_libs-0.1.6.post1.dist-info}/top_level.txt

@@ -1,3 +1,5 @@
+CI
 evaluation
 gr_libs
+tests
 tutorials

tests/test_graml.py

@@ -0,0 +1,16 @@
+from tutorials.graml_minigrid_tutorial import run_graml_minigrid_tutorial
+from tutorials.graml_panda_tutorial import run_graml_panda_tutorial
+from tutorials.graml_parking_tutorial import run_graml_parking_tutorial
+from tutorials.graml_point_maze_tutorial import run_graml_point_maze_tutorial
+
+def test_graml_minigrid_tutorial():
+	run_graml_minigrid_tutorial()
+
+def test_graml_panda_tutorial():
+	run_graml_panda_tutorial()
+
+def test_graml_parking_tutorial():
+	run_graml_parking_tutorial()
+
+def test_graml_point_maze_tutorial():
+	run_graml_point_maze_tutorial()

tests/test_graql.py

@@ -0,0 +1,4 @@
+from tutorials.graql_minigrid_tutorial import run_graql_minigrid_tutorial
+
+def test_graql_minigrid_tutorial():
+	run_graql_minigrid_tutorial()

tutorials/graml_minigrid_tutorial.py

@@ -4,27 +4,31 @@ from gr_libs.ml.tabular.tabular_q_learner import TabularQLearner
 from gr_libs.ml.utils.format import random_subset_with_order
 from gr_libs import ExpertBasedGraml
 
-recognizer = ExpertBasedGraml(
-	domain_name=MINIGRID,
-	env_name="MiniGrid-SimpleCrossingS13N4"
-)
+def run_graml_minigrid_tutorial():
+    recognizer = ExpertBasedGraml(
+        domain_name=MINIGRID,
+        env_name="MiniGrid-SimpleCrossingS13N4"
+    )
 
-recognizer.domain_learning_phase(base_goals=[(11,1), (11,11), (1,11), (7,11), (8,1), (10,6), (6,9), (11,3), (11,5)],
-                                 train_configs=[(QLEARNING, 100000) for _ in range(9)])
+    recognizer.domain_learning_phase(base_goals=[(11,1), (11,11), (1,11), (7,11), (8,1), (10,6), (6,9), (11,3), (11,5)],
+                                    train_configs=[(QLEARNING, 100000) for _ in range(9)])
 
-recognizer.goals_adaptation_phase(
-    dynamic_goals = [(11,1), (11,11), (1,11)],
-    dynamic_train_configs=[(QLEARNING, 100000) for _ in range(3)] # for expert sequence generation.
-)
-# TD3 is different from recognizer and expert algorithms, which are SAC #
-actor = TabularQLearner(domain_name="minigrid", problem_name="MiniGrid-SimpleCrossingS13N4-DynamicGoal-11x1-v0", algorithm=QLEARNING, num_timesteps=100000)
-actor.learn()
-# sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
-full_sequence = actor.generate_observation(
-    action_selection_method=stochastic_amplified_selection,
-    random_optimalism=True, # the noise that's added to the actions
-)
+    recognizer.goals_adaptation_phase(
+        dynamic_goals = [(11,1), (11,11), (1,11)],
+        dynamic_train_configs=[(QLEARNING, 100000) for _ in range(3)] # for expert sequence generation.
+    )
+    # TD3 is different from recognizer and expert algorithms, which are SAC #
+    actor = TabularQLearner(domain_name="minigrid", problem_name="MiniGrid-SimpleCrossingS13N4-DynamicGoal-11x1-v0", algorithm=QLEARNING, num_timesteps=100000)
+    actor.learn()
+    # sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
+    full_sequence = actor.generate_observation(
+        action_selection_method=stochastic_amplified_selection,
+        random_optimalism=True, # the noise that's added to the actions
+    )
 
-partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
-closest_goal = recognizer.inference_phase(partial_sequence, (11,1), 0.5)
-print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: (11, 1)")
+    partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
+    closest_goal = recognizer.inference_phase(partial_sequence, (11,1), 0.5)
+    print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: (11, 1)")
+
+if __name__ == "__main__":
+    run_graml_minigrid_tutorial()

tutorials/graml_panda_tutorial.py

@@ -9,29 +9,33 @@ from gr_libs.ml.neural.deep_rl_learner import DeepRLAgent, GCDeepRLAgent
 from gr_libs.ml.utils.format import random_subset_with_order
 from gr_libs import GCGraml
 
-recognizer = GCGraml( # TODO make these tutorials into pytests
-    domain_name=PANDA,
-	env_name="PandaMyReachDense"
-)
-recognizer.domain_learning_phase(
-	base_goals=[np.array([PandaProperty.sample_goal()]) for _ in range(1,30)],
-	train_configs=[(SAC, 800000)]
-)
-recognizer.goals_adaptation_phase(
-	dynamic_goals=[np.array([[-0.1, -0.1, 0.1]]), np.array([[-0.1, 0.1, 0.1]]), np.array([[0.2, 0.2, 0.1]])]
-)
-# TD3 is different from recognizer and expert algorithms, which are SAC #
-property_type = domain_to_env_property(PANDA)
-env_property = property_type("PandaMyReachDense")
-problem_name = env_property.goal_to_problem_str(np.array([[-0.1, -0.1, 0.1]]))
-actor = DeepRLAgent(domain_name=PANDA, problem_name=problem_name, algorithm=PPO, num_timesteps=400000)
-actor.learn()
-# sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
-full_sequence = actor.generate_observation(
-    action_selection_method=stochastic_amplified_selection,
-    random_optimalism=True, # the noise that's added to the actions
-)
+def run_graml_panda_tutorial():
+    recognizer = GCGraml( # TODO make these tutorials into pytests
+        domain_name=PANDA,
+        env_name="PandaMyReachDense"
+    )
+    recognizer.domain_learning_phase(
+        base_goals=[np.array([PandaProperty.sample_goal()]) for _ in range(1,30)],
+        train_configs=[(SAC, 800000)]
+    )
+    recognizer.goals_adaptation_phase(
+        dynamic_goals=[np.array([[-0.1, -0.1, 0.1]]), np.array([[-0.1, 0.1, 0.1]]), np.array([[0.2, 0.2, 0.1]])]
+    )
+    # TD3 is different from recognizer and expert algorithms, which are SAC #
+    property_type = domain_to_env_property(PANDA)
+    env_property = property_type("PandaMyReachDense")
+    problem_name = env_property.goal_to_problem_str(np.array([[-0.1, -0.1, 0.1]]))
+    actor = DeepRLAgent(domain_name=PANDA, problem_name=problem_name, algorithm=PPO, num_timesteps=400000)
+    actor.learn()
+    # sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
+    full_sequence = actor.generate_observation(
+        action_selection_method=stochastic_amplified_selection,
+        random_optimalism=True, # the noise that's added to the actions
+    )
 
-partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
-closest_goal = recognizer.inference_phase(partial_sequence, np.array([[-0.1, -0.1, 0.1]]), 0.5)
-print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: [-0.1, -0.1, 0.1]")
+    partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
+    closest_goal = recognizer.inference_phase(partial_sequence, np.array([[-0.1, -0.1, 0.1]]), 0.5)
+    print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: [-0.1, -0.1, 0.1]")
+
+if __name__ == "__main__":
+    run_graml_panda_tutorial()

tutorials/graml_parking_tutorial.py

@@ -6,29 +6,33 @@ from gr_libs.ml.neural.deep_rl_learner import DeepRLAgent, GCDeepRLAgent
 from gr_libs.ml.utils.format import random_subset_with_order
 from gr_libs.recognizer.graml.graml_recognizer import ExpertBasedGraml, GCGraml
 
-recognizer = GCGraml(
-	domain_name=PARKING,
-	env_name="Parking-S-14-PC-"
-)
-					
-recognizer.domain_learning_phase(
-	[i for i in range(1,21)],
-	[(PPO, 200000)]
-)
-recognizer.goals_adaptation_phase(
-    dynamic_goals = ["1", "11", "21"]
-	# no need for expert sequence generation since GCRL is used
-)
+def run_graml_parking_tutorial():
+    recognizer = GCGraml(
+        domain_name=PARKING,
+        env_name="Parking-S-14-PC-"
+    )
+                        
+    recognizer.domain_learning_phase(
+        [i for i in range(1,21)],
+        [(PPO, 200000)]
+    )
+    recognizer.goals_adaptation_phase(
+        dynamic_goals = ["1", "11", "21"]
+        # no need for expert sequence generation since GCRL is used
+    )
 
-# TD3 is different from recognizer and expert algorithms, which are SAC #
-actor = DeepRLAgent(domain_name="parking", problem_name="Parking-S-14-PC--GI-11-v0", algorithm=TD3, num_timesteps=400000)
-actor.learn()
-# sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
-full_sequence = actor.generate_observation(
-    action_selection_method=stochastic_amplified_selection,
-    random_optimalism=True, # the noise that's added to the actions
-)
+    # TD3 is different from recognizer and expert algorithms, which are SAC #
+    actor = DeepRLAgent(domain_name="parking", problem_name="Parking-S-14-PC--GI-11-v0", algorithm=TD3, num_timesteps=400000)
+    actor.learn()
+    # sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
+    full_sequence = actor.generate_observation(
+        action_selection_method=stochastic_amplified_selection,
+        random_optimalism=True, # the noise that's added to the actions
+    )
 
-partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
-closest_goal = recognizer.inference_phase(partial_sequence, ParkingProperty("Parking-S-14-PC--GI-11-v0").str_to_goal(), 0.5)
-print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: 11")
+    partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
+    closest_goal = recognizer.inference_phase(partial_sequence, ParkingProperty("Parking-S-14-PC--GI-11-v0").str_to_goal(), 0.5)
+    print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: 11")
+
+if __name__ == "__main__":
+    run_graml_parking_tutorial()

tutorials/graml_point_maze_tutorial.py

@@ -6,30 +6,34 @@ from gr_libs.ml.neural.deep_rl_learner import DeepRLAgent
 from gr_libs.ml.utils.format import random_subset_with_order
 from gr_libs.recognizer.graml.graml_recognizer import ExpertBasedGraml
 
-recognizer = ExpertBasedGraml(
-	domain_name=POINT_MAZE,
-	env_name="PointMaze-FourRoomsEnvDense-11x11"
-)
+def run_graml_point_maze_tutorial():
+    recognizer = ExpertBasedGraml(
+        domain_name=POINT_MAZE,
+        env_name="PointMaze-FourRoomsEnvDense-11x11"
+    )
 
-recognizer.domain_learning_phase(
-	[(9,1), (9,9), (1,9), (3,3), (3,4), (8,2), (3,7), (2,8)],
-	[(SAC, 200000) for _ in range(8)]
-)
+    recognizer.domain_learning_phase(
+        [(9,1), (9,9), (1,9), (3,3), (3,4), (8,2), (3,7), (2,8)],
+        [(SAC, 200000) for _ in range(8)]
+    )
 
-recognizer.goals_adaptation_phase(
-    dynamic_goals = [(4,4), (7,3), (3,7)],
-    dynamic_train_configs=[(SAC, 200000) for _ in range(3)] # for expert sequence generation.
-)
+    recognizer.goals_adaptation_phase(
+        dynamic_goals = [(4,4), (7,3), (3,7)],
+        dynamic_train_configs=[(SAC, 200000) for _ in range(3)] # for expert sequence generation.
+    )
 
-# TD3 is different from recognizer and expert algorithms, which are SAC #
-actor = DeepRLAgent(domain_name="point_maze", problem_name="PointMaze-FourRoomsEnvDense-11x11-Goal-4x4", algorithm=TD3, num_timesteps=200000)
-actor.learn()
-# sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
-full_sequence = actor.generate_observation(
-    action_selection_method=stochastic_amplified_selection,
-    random_optimalism=True, # the noise that's added to the actions
-)
+    # TD3 is different from recognizer and expert algorithms, which are SAC #
+    actor = DeepRLAgent(domain_name="point_maze", problem_name="PointMaze-FourRoomsEnvDense-11x11-Goal-4x4", algorithm=TD3, num_timesteps=200000)
+    actor.learn()
+    # sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
+    full_sequence = actor.generate_observation(
+        action_selection_method=stochastic_amplified_selection,
+        random_optimalism=True, # the noise that's added to the actions
+    )
 
-partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)))
-closest_goal = recognizer.inference_phase(partial_sequence, PointMazeProperty("PointMaze-FourRoomsEnvDense-11x11-Goal-4x4").str_to_goal(), 0.5)
-print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: (4, 4)")
+    partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)))
+    closest_goal = recognizer.inference_phase(partial_sequence, PointMazeProperty("PointMaze-FourRoomsEnvDense-11x11-Goal-4x4").str_to_goal(), 0.5)
+    print(f"closest_goal returned by GRAML: {closest_goal}\nactual goal actor aimed towards: (4, 4)")
+
+if __name__ == "__main__":
+    run_graml_point_maze_tutorial()

tutorials/graql_minigrid_tutorial.py

@@ -4,26 +4,31 @@ from gr_libs.ml.tabular.tabular_q_learner import TabularQLearner
 from gr_libs.ml.utils.format import random_subset_with_order
 from gr_libs import Graql
 
-recognizer = Graql(
-	domain_name="minigrid",
-	env_name="MiniGrid-SimpleCrossingS13N4"
-)
+def run_graql_minigrid_tutorial():
+    recognizer = Graql(
+        domain_name="minigrid",
+        env_name="MiniGrid-SimpleCrossingS13N4"
+    )
 
-#Graql doesn't have a domain learning phase, so we skip it
+    #Graql doesn't have a domain learning phase, so we skip it
 
-recognizer.goals_adaptation_phase(
-    dynamic_goals = [(11,1), (11,11), (1,11)],
-    dynamic_train_configs=[(QLEARNING, 100000) for _ in range(3)] # for expert sequence generation.
-)
-# TD3 is different from recognizer and expert algorithms, which are SAC #
-actor = TabularQLearner(domain_name="minigrid", problem_name="MiniGrid-SimpleCrossingS13N4-DynamicGoal-11x1-v0", algorithm=QLEARNING, num_timesteps=100000)
-actor.learn()
-# sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
-full_sequence = actor.generate_observation(
-    action_selection_method=stochastic_amplified_selection,
-    random_optimalism=True, # the noise that's added to the actions
-)
+    recognizer.goals_adaptation_phase(
+        dynamic_goals = [(11,1), (11,11), (1,11)],
+        dynamic_train_configs=[(QLEARNING, 100000) for _ in range(3)] # for expert sequence generation.
+    )
+    # TD3 is different from recognizer and expert algorithms, which are SAC #
+    actor = TabularQLearner(domain_name="minigrid", problem_name="MiniGrid-SimpleCrossingS13N4-DynamicGoal-11x1-v0", algorithm=QLEARNING, num_timesteps=100000)
+    actor.learn()
+    # sample is generated stochastically to simulate suboptimal behavior, noise is added to the actions values #
+    full_sequence = actor.generate_observation(
+        action_selection_method=stochastic_amplified_selection,
+        random_optimalism=True, # the noise that's added to the actions
+    )
 
-partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
-closest_goal = recognizer.inference_phase(partial_sequence, (11,1), 0.5)
-print(f"closest_goal returned by Graql: {closest_goal}\nactual goal actor aimed towards: (11, 1)")
+    partial_sequence = random_subset_with_order(full_sequence, (int)(0.5 * len(full_sequence)), is_consecutive=False)
+    closest_goal = recognizer.inference_phase(partial_sequence, (11,1), 0.5)
+    print(f"closest_goal returned by Graql: {closest_goal}\nactual goal actor aimed towards: (11, 1)")
+    return closest_goal, (11,1)
+
+if __name__ == "__main__":
+    run_graql_minigrid_tutorial()