firs commit for our new MARL algorithms library, contains working implementations of IAC, SNAC and SEAC

environments/factory/__init__.py

@@ -1,22 +1,25 @@
-def make(env_name, pomdp_r=2, max_steps=400, stack_n_frames=3, n_agents=1,  individual_rewards=False):
+def make(env_name, pomdp_r=2, max_steps=400, stack_n_frames=3, n_agents=1, individual_rewards=False):
     import yaml
     from pathlib import Path
     from environments.factory.combined_factories import DirtItemFactory
     from environments.factory.factory_item import ItemFactory, ItemProperties
-    from environments.factory.factory_dirt import DirtProperties, DirtFactory
-    from environments.utility_classes import MovementProperties, ObservationProperties, AgentRenderOptions
+    from environments.factory.factory_dirt import DirtProperties, DirtFactory, RewardsDirt
+    from environments.utility_classes import AgentRenderOptions
 
     with (Path(__file__).parent / 'levels' / 'parameters' / f'{env_name}.yaml').open('r') as stream:
         dictionary = yaml.load(stream, Loader=yaml.FullLoader)
 
-    obs_props = ObservationProperties(render_agents=AgentRenderOptions.COMBINED,
-                                      frames_to_stack=stack_n_frames, pomdp_r=pomdp_r)
+    obs_props = dict(render_agents=AgentRenderOptions.COMBINED,
+                     pomdp_r=pomdp_r,
+                     indicate_door_area=True,
+                     show_global_position_info=False,
+                     frames_to_stack=stack_n_frames)
 
-    factory_kwargs = dict(n_agents=n_agents, individual_rewards=individual_rewards,
-                          max_steps=max_steps, obs_prop=obs_props,
-                          mv_prop=MovementProperties(**dictionary['movement_props']),
-                          dirt_prop=DirtProperties(**dictionary['dirt_props']),
-                          record_episodes=False, verbose=False, **dictionary['factory_props']
+    factory_kwargs = dict(**dictionary,
+                          n_agents=n_agents,
+                          individual_rewards=individual_rewards,
+                          max_steps=max_steps,
+                          obs_prop=obs_props,
+                          verbose=False,
                           )
-
     return DirtFactory(**factory_kwargs).__enter__()

environments/factory/levels/parameters/DirtyFactory-v0.yaml

@@ -1,8 +1,12 @@
-movement_props:
+parse_doors:                True
+doors_have_area:            True
+done_at_collision:          False
+level_name:                 "rooms"
+mv_prop:
     allow_diagonal_movement:    True
     allow_square_movement:      True
     allow_no_op:                False
-dirt_props:
+dirt_prop:
     initial_dirt_ratio:         0.35
     initial_dirt_spawn_r_var :  0.1
     clean_amount:               0.34
@@ -12,8 +16,15 @@ dirt_props:
     spawn_frequency:            0
     max_spawn_ratio:            0.05
     dirt_smear_amount:          0.0
-    agent_can_interact:         True
-factory_props:
-    parse_doors:                True
-    level_name:                 "rooms"
-    doors_have_area:            False
+    done_when_clean:            True
+rewards_base:
+    MOVEMENTS_VALID:    0
+    MOVEMENTS_FAIL:     0
+    NOOP:               0
+    USE_DOOR_VALID:     0
+    USE_DOOR_FAIL:      0
+    COLLISION:          0
+rewards_dirt:
+    CLEAN_UP_VALID:       1
+    CLEAN_UP_FAIL:        0
+    CLEAN_UP_LAST_PIECE:  5