Stochastic actions

enums/action_probabilities.py

@@ -0,0 +1,15 @@
+# Note: This really isn't an enum but a constant. Should this be placed in a different folder?
+from enums.player_action import PlayerAction
+
+Action_Probabilities = {
+    PlayerAction.NOP : {PlayerAction.NOP: 1},
+    PlayerAction.NORTH : {PlayerAction.NORTH: 1, PlayerAction.SOUTH: 0, PlayerAction.EAST: 0, PlayerAction.WEST: 0, PlayerAction.NOP: 0},
+    PlayerAction.SOUTH : {PlayerAction.NORTH: 0, PlayerAction.SOUTH: 1, PlayerAction.EAST: 0, PlayerAction.WEST: 0, PlayerAction.NOP: 0},
+    PlayerAction.EAST : {PlayerAction.NORTH: 0, PlayerAction.SOUTH: 0, PlayerAction.EAST: 1, PlayerAction.WEST: 0, PlayerAction.NOP: 0},
+    PlayerAction.WEST : {PlayerAction.NORTH: 0, PlayerAction.SOUTH: 0, PlayerAction.EAST: 0, PlayerAction.WEST: 1, PlayerAction.NOP: 0},
+    PlayerAction.INTERACT : {PlayerAction.INTERACT: 1, PlayerAction.NOP: 0},
+    PlayerAction.TOGGLE : {PlayerAction.TOGGLE: 1, PlayerAction.NOP: 0},
+    PlayerAction.CANCEL : {PlayerAction.CANCEL: 1, PlayerAction.NOP: 0},
+    PlayerAction.PICKUP : {PlayerAction.PICKUP: 1, PlayerAction.NOP: 0},
+    PlayerAction.RESET : {PlayerAction.RESET: 1},
+}

enums/player_action.py

@@ -12,3 +12,16 @@ class PlayerAction(IntEnum):
     CANCEL = 7,
     PICKUP = 8,
     RESET = 9,
+
+PlayerActionTable = {
+    "NOP" : PlayerAction.NOP,
+    "NORTH" : PlayerAction.NORTH,
+    "SOUTH" : PlayerAction.SOUTH,
+    "EAST" : PlayerAction.EAST,
+    "WEST" : PlayerAction.WEST,
+    "INTERACT" : PlayerAction.INTERACT,
+    "TOGGLE" : PlayerAction.TOGGLE,
+    "CANCEL" : PlayerAction.CANCEL,
+    "PICKUP" : PlayerAction.PICKUP,
+    "RESET" : PlayerAction.RESET,
+}

env.py

@@ -1,7 +1,9 @@
+import random
 import time
 
+from enums.action_probabilities import Action_Probabilities
 import gymnasium as gym
-from enums.player_action import PlayerAction
+from enums.player_action import PlayerAction, PlayerActionTable
 from game import Game
 
 MOVEMENT_ACTIONS = [PlayerAction.NORTH, PlayerAction.SOUTH, PlayerAction.EAST, PlayerAction.WEST]
@@ -11,7 +13,7 @@ class SupermarketEnv(gym.Env):
 
     def __init__(self, num_players=1, player_speed=0.15, keyboard_input=False, render_messages=True, bagging=False,
                  headless=False, initial_state_filename=None, follow_player=-1, random_start=False,
-                 render_number=False, max_num_items=33, player_sprites=None, record_path=None, stay_alive=False):
+                 render_number=False, max_num_items=33, player_sprites=None, record_path=None, stay_alive=False, stochastic=False):
 
         super(SupermarketEnv, self).__init__()
 
@@ -40,11 +42,27 @@ def __init__(self, num_players=1, player_speed=0.15, keyboard_input=False, rende
         self.observation_space = gym.spaces.Dict()
         self.headless = headless
         self.random_start = random_start
+        self.action_probability = {}
+
+        if (stochastic): # storing probability of action success rate
+            filename = input("Input stochastic probability file name: ")
+            with open(filename, "r") as file: 
+                content = file.read()
+                for row in content.split("\n"):
+                    action_row = list(map(lambda column: column.strip(": "), row.split("\t")))
+                    probability_pairs = map(lambda result: tuple(result.split(" ")), action_row[1:])
+                    self.action_probability[PlayerActionTable[action_row[0]]] = dict(map(lambda pair: (PlayerActionTable[pair[0]], int(pair[1])), probability_pairs))
+        else: 
+            self.action_probability = Action_Probabilities
+
+    def get_stochastic_action(self, action):
+        return random.choices(list(self.action_probability[action].keys()), weights=list(self.action_probability[action].values()), k=1)[0]
 
     def step(self, action):
         done = False
         for i, player_action in enumerate(action):
             player_action, arg = player_action
+            player_action = self.get_stochastic_action(player_action)
             if player_action in MOVEMENT_ACTIONS:
                 self.unwrapped.game.player_move(i, player_action)
             elif player_action == PlayerAction.NOP:
@@ -114,6 +132,7 @@ def convert_action(self, player_action):
     def step(self, player_action):
         done = False
         i, player_action, arg = player_action
+        player_action = self.get_stochastic_action(player_action)
         if player_action in MOVEMENT_ACTIONS:
             self.unwrapped.game.player_move(i, player_action)
         elif player_action == PlayerAction.NOP:

socket_env.py

@@ -259,6 +259,13 @@ def accept_wrapper(sock):
         '--stay_alive',
         action='store_true',
     )
+
+    parser.add_argument(
+        '--stochastic',
+        help="file to read stochastic success probabilities of actions",
+        action='store_true',
+    )
+
 
     args = parser.parse_args()
 
@@ -276,7 +283,8 @@ def accept_wrapper(sock):
                          render_number=args.render_number,
                          player_sprites=args.player_sprites,
                          record_path=args.record_path,
-                         stay_alive=args.stay_alive
+                         stay_alive=args.stay_alive, 
+                         stochastic = args.stochastic
                          )
 
     norms = [CartTheftNorm(),

stochastic_probability.txt

@@ -0,0 +1,10 @@
+NOP:	NOP 1
+NORTH:	NORTH 1	SOUTH 0	EAST 0	WEST 0	NOP 0
+SOUTH:	NORTH 0	SOUTH 1	EAST 0	WEST 0	NOP 0
+EAST:	NORTH 0	SOUTH 0	EAST 1	WEST 0	NOP 0
+WEST:	NORTH 0	SOUTH 0	EAST 0	WEST 1	NOP 0
+INTERACT:	INTERACT 1	NOP 0
+TOGGLE:	 TOGGLE 1	NOP 0
+CANCEL:	 CANCEL 1	NOP 0
+PICKUP:	 PICKUP 1	NOP 0
+RESET:	RESET 1