def rule_based_action(self, player):
from kaggle_environments.envs.hungry_geese.hungry_geese import (
Observation,
Configuration,
Action,
GreedyAgent,
)
action_map = {
"N": Action.NORTH,
"S": Action.SOUTH,
"W": Action.WEST,
"E": Action.EAST,
}
agent = GreedyAgent(Configuration({"rows": 7, "columns": 11}))
agent.last_action = (
action_map[self.ACTION[self.last_actions[player]][0]]
if player in self.last_actions
else None
)
obs = {
**self.obs_list[-1][0]["observation"],
**self.obs_list[-1][player]["observation"],
}
action = agent(Observation(obs))
return self.ACTION.index(action)
def agent(obs_dict, config_dict):
global ralph_last_action
#np.set_printoptions(precision=2)
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
astar = AStar(observation, configuration)
path = astar.search()
choice = "WEST"
if (path is not None):
cam = []
if (len(path.caminho) > 1):
cam = path.caminho[1]
else:
cam = path.next_p
choice = get_choice(cam, astar.start, astar.rows, astar.cols)
print(choice)
'''for line in astar.heu_map:
print(line)'''
else:
if (ralph_last_action[observation.index] is not None):
choice = opposite(
get_choice(ralph_last_action[observation.index], astar.start,
astar.rows, astar.cols))
else:
coice = "WEST"
print()
print()
print()
ralph_last_action[observation.index] = astar.start
return choice
def agent(obs, conf):
global directions
obs = Observation(obs)
conf = Configuration(conf)
board = np.zeros((7, 11), dtype=int)
#Obstacle-ize your opponents
for ind, goose in enumerate(obs.geese):
if ind == obs.index or len(goose) == 0:
continue
for direction in range(4):
moved = move(goose, direction)
for part in moved:
board[part//11][part%11] -= 1
#Obstacle-ize your body, except the last part
if len(obs.geese[obs.index]) > 1:
for k in obs.geese[obs.index][:-1]:
board[k//11][k%11] -= 1
#Count food only if there's no chance an opponent will meet you there
for f in obs.food:
board[f//11][f%11] += (board[f//11][f%11] == 0)
return directions[greedy_choose(obs.geese[obs.index][0], board)]
def process_state(self, obs_dict, config_dict):
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
b_state={
'blank': 0, # Empty cell
'x': -1, # Head of opponent player
'h': 1, # Head of current player
'b': -2, # Body cell
'f': 2, # Food
}
# строим доску
board = b_state['blank'] * np.zeros(N_CELL).astype(int).reshape((N_ROWS, N_COLS))
# гуси
geese = observation.geese
for goose in geese:
for i, cell in enumerate(goose):
row, col = row_col(cell, configuration.columns)
board[row][col] = b_state['b']
for i in range(len(geese)):
if(len(geese[i]) <= 0):
continue
head_row, head_col = row_col(geese[i][0], configuration.columns)
if player_index == i:
board[head_row][head_col] = b_state['h']
else:
board[head_row][head_col] = b_state['x']
# food
foods = observation.food
for food in foods:
food_row, food_col = row_col(food, configuration.columns)
board[food_row][food_col] = b_state['f']
board = self.centroid_agent(board, geese[player_index][0], configuration)
return board
def rule_based_action(self, player):
from kaggle_environments.envs.hungry_geese.hungry_geese import Observation, Configuration, Action, GreedyAgent
action_map = {'N': Action.NORTH, 'S': Action.SOUTH, 'W': Action.WEST, 'E': Action.EAST}
agent = GreedyAgent(Configuration({'rows': 7, 'columns': 11}))
agent.last_action = action_map[self.ACTION[self.last_actions[player]][0]] if player in self.last_actions else None
obs = {**self.obs_list[-1][0]['observation'], **self.obs_list[-1][player]['observation']}
action = agent(Observation(obs))
return self.ACTION.index(action)
def agent(observation, configuration):
observation = Observation(observation)
configuration = Configuration(configuration)
rows, columns = configuration.rows, configuration.columns
food = observation.food
geese = observation.geese
opponents = [
goose
for index, goose in enumerate(geese)
if index != observation.index and len(goose) > 0
]
# Don't move adjacent to any heads
head_adjacent_positions = {
opponent_head_adjacent
for opponent in opponents
for opponent_head in [opponent[0]]
for opponent_head_adjacent in adjacent_positions(opponent_head, rows, columns)
}
# Don't move into any bodies
bodies = {position for goose in geese for position in goose[0:-1]}
# Don't move into tails of heads that are adjacent to food
tails = {
opponent[-1]
for opponent in opponents
for opponent_head in [opponent[0]]
if any(
adjacent_position in food
# Head of opponent is adjacent to food so tail is not safe
for adjacent_position in adjacent_positions(opponent_head, rows, columns)
)
}
# Move to the closest food
position = geese[observation.index][0]
actions = {
action: min_distance(new_position, food, columns)
for action in Action
for new_position in [translate(position, action, columns, rows)]
if (
new_position not in head_adjacent_positions and
new_position not in bodies and
new_position not in tails
)
}
if any(actions):
return min(actions, key=actions.get).name
return random_agent()
def agent_ORIGINAL(obs_dict, config_dict):
"""This agent always moves toward observation.food[0] but does not take advantage of board wrapping"""
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
player_row, player_column = row_col(player_head, configuration.columns)
food = observation.food[0]
food_row, food_column = row_col(food, configuration.columns)
if food_row > player_row:
return Action.SOUTH.name
if food_row < player_row:
return Action.NORTH.name
if food_column > player_column:
return Action.EAST.name
return Action.WEST.name
def __init__(self,
model,
nchannel=3,
ncheck=10,
ndisplay=3.14,
n_smart=2,
rule_bots=[],
switch_epoch=500,
border=0,
load_weights=None,
shuffle=False,
save_epochs=100):
self.env = make('hungry_geese')
self.env.reset(num_agents=4)
self.n_epochs = 1000
self.epoch = 1
self.ncheck = ncheck
self.ndisplay = ndisplay
self.nsmart = n_smart
self.conf = Configuration(self.env.configuration)
self.actions = np.array(['NORTH', 'SOUTH', 'WEST', 'EAST'])
self.step_history = np.zeros(ncheck).astype(int)
self.goose_steps = [0, 0, 0, 0]
self.goose = model
self.rule_bots = [rb(self.conf) for rb in rule_bots]
self.border = border
self.shuffle = shuffle
self.save_epochs = save_epochs
self.nchannel = nchannel
self.first = True
if load_weights != None:
print(f'Loading weights: {load_weights}')
self.goose.epsilon = 0.01
self.goose.Q.model.load_weights(load_weights)
self.goose.V_eval.model.load_weights(load_weights)
self.goose.V_targ.model.load_weights(load_weights)
self.prev_length = [1, 1, 1, 1]
self.prev_act = ['ACTIVE', 'ACTIVE', 'ACTIVE', 'ACTIVE']
self.prev_action = [None, None, None, None]
self.best = 0
def __init__(self, env=None, rows=7, columns=11, hunger_rate=40, min_food=2, num_agents=4):
if env is not None:
self.env = env.clone()
else:
self.env = make(
"hungry_geese",
configuration={
"rows": rows,
"columns": columns,
"hunger_rate": hunger_rate,
"min_food": min_food
},
debug=False
)
self.config = Configuration(self.env.configuration)
self.num_agents = num_agents
self.action_size = len(Action)
self.str_to_action = {'NORTH': Action.NORTH, 'EAST': Action.EAST, 'SOUTH': Action.SOUTH, 'WEST': Action.WEST}
self.action_to_num = {'NORTH': 1, 'EAST': 2, 'SOUTH': 3, 'WEST': 4}
self.prev_actions = None
def processObservations(self, obs_dict, config_dict, alive=True):
if not alive: return self._state.EmptyObservations
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
if self._agent is None:
self._agent = GreedyAgent(configuration)
state, validAct, actionsMapping = self._state.local(observation, configuration)
self._agent.last_action = Action[self._prevAction] # sync
self._actName = self._agent(observation)
if self._actName not in actionsMapping:
self._actName = next(
(act for i, act in enumerate(actionsMapping) if 0 < validAct[i]),
actionsMapping[1]
)
self._actID = actionsMapping.index(self._actName)
return state
def preprocess_env(self, observation, configuration):
observation = Observation(observation)
configuration = Configuration(configuration)
self.rows, self.columns = configuration.rows, configuration.columns
self.my_index = observation.index
self.hunger_rate = configuration.hunger_rate
self.min_food = configuration.min_food
self.my_head, self.my_tail = observation.geese[
self.my_index][0], observation.geese[self.my_index][-1]
self.my_body = [pos for pos in observation.geese[self.my_index][1:-1]]
self.geese = [
g for i, g in enumerate(observation.geese)
if i != self.my_index and len(g) > 0
]
self.geese_cells = [pos for g in self.geese for pos in g if len(g) > 0]
self.occupied = [p for p in self.geese_cells]
self.occupied.extend([p for p in observation.geese[self.my_index]])
self.heads = [
g[0] for i, g in enumerate(observation.geese)
if i != self.my_index and len(g) > 0
]
self.bodies = [
pos for i, g in enumerate(observation.geese) for pos in g[1:-1]
if i != self.my_index and len(g) > 2
]
self.tails = [
g[-1] for i, g in enumerate(observation.geese)
if i != self.my_index and len(g) > 1
]
self.food = [f for f in observation.food]
self.adjacent_to_heads = [
pos for head in self.heads
for pos in self._adjacent_positions(head)
]
self.adjacent_to_bodies = [
pos for body in self.bodies
for pos in self._adjacent_positions(body)
]
self.adjacent_to_tails = [
pos for tail in self.tails
for pos in self._adjacent_positions(tail)
]
self.adjacent_to_geese = self.adjacent_to_heads + self.adjacent_to_bodies
self.danger_zone = self.adjacent_to_geese
#Cell occupation
self.cell_states = [
CellState.EMPTY.value for _ in range(self.rows * self.columns)
]
for g in self.geese:
for pos in g:
self.cell_states[pos] = CellState.GOOSE.value
for pos in self.heads:
self.cell_states[pos] = CellState.GOOSE.value
for pos in self.my_body:
self.cell_states[pos] = CellState.GOOSE.value
#detect dead-ends
self.dead_ends = []
for pos_i, _ in enumerate(self.cell_states):
if self.cell_states[pos_i] != CellState.EMPTY.value:
continue
adjacent = self._adjacent_positions(pos_i)
adjacent_states = [
self.cell_states[adj_pos] for adj_pos in adjacent
if adj_pos != self.my_head
]
num_blocked = sum(adjacent_states)
if num_blocked >= (CellState.GOOSE.value * 3):
self.dead_ends.append(pos_i)
#check for extended dead-ends
new_dead_ends = [pos for pos in self.dead_ends]
while new_dead_ends != []:
for pos in new_dead_ends:
self.cell_states[pos] = CellState.GOOSE.value
self.dead_ends.append(pos)
new_dead_ends = []
for pos_i, _ in enumerate(self.cell_states):
if self.cell_states[pos_i] != CellState.EMPTY.value:
continue
adjacent = self._adjacent_positions(pos_i)
adjacent_states = [
self.cell_states[adj_pos] for adj_pos in adjacent
if adj_pos != self.my_head
]
num_blocked = sum(adjacent_states)
if num_blocked >= (CellState.GOOSE.value * 3):
new_dead_ends.append(pos_i)
def straightforward_bfs(obs_dict, config_dict):
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
start_row, start_col = row_col(player_head, configuration.columns)
mask = np.zeros((configuration.rows, configuration.columns))
for current_id in range(4):
current_goose = observation.geese[current_id]
for block in current_goose:
current_row, current_col = row_col(block, configuration.columns)
mask[current_row, current_col] = -1
food_coords = []
for food_id in range(configuration.min_food):
food = observation.food[food_id]
current_row, current_col = row_col(food, configuration.columns)
mask[current_row, current_col] = 2
food_coords.append((current_row, current_col))
last_action = bfs(start_row, start_col, mask, food_coords)
global LAST_ACTION
up_x = start_row + 1 if start_row != 6 else 0
down_x = start_row - 1 if start_row != 0 else 6
left_y = start_col - 1 if start_col != 0 else 10
right_y = start_col + 1 if start_col != 10 else 0
step = Action.NORTH.name
if last_action == 0:
step = Action.SOUTH.name
if LAST_ACTION == Action.NORTH.name:
if mask[down_x, start_col] != -1:
step = Action.NORTH.name
elif mask[start_row, left_y] != -1:
step = Action.WEST.name
elif mask[start_row, right_y] != -1:
step = Action.EAST.name
if last_action == 1:
step = Action.NORTH.name
if LAST_ACTION == Action.SOUTH.name:
if mask[up_x, start_col] != -1:
step = Action.SOUTH.name
elif mask[start_row, left_y] != -1:
step = Action.WEST.name
elif mask[start_row, right_y] != -1:
step = Action.EAST.name
if last_action == 2:
step = Action.WEST.name
if LAST_ACTION == Action.EAST.name:
if mask[up_x, start_col] != -1:
step = Action.SOUTH.name
elif mask[down_x, start_col] != -1:
step = Action.NORTH.name
elif mask[start_row, right_y] != -1:
step = Action.EAST.name
if last_action == 3:
step = Action.EAST.name
if LAST_ACTION == Action.WEST.name:
if mask[up_x, start_col] != -1:
step = Action.SOUTH.name
elif mask[down_x, start_col] != -1:
step = Action.NORTH.name
elif mask[start_row, left_y] != -1:
step = Action.WEST.name
LAST_ACTION = step
return step
def agent(obs_dict, config_dict):
global last_move
global last_eaten
global last_size
global step
#print ("==============================================")
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
player_row, player_column = row_col(player_head, configuration.columns)
if (len(player_goose) > last_size):
last_size = len(player_goose)
last_eaten = step
step += 1
moves = {1: 'SOUTH', 2: 'NORTH', 3: 'EAST', 4: 'WEST'}
board = np.zeros((7, 11))
# Adding food to board
for food in observation.food:
x, y = row_col(food, configuration.columns)
#print ("Food cell on ({}, {})".format(x, y))
board[x, y] = FOOD_CELL
# Adding geese to the board
for i in range(4):
goose = observation.geese[i]
# Skip if goose is dead
if len(goose) == 0:
continue
# If it's an opponent
if i != player_index:
x, y = row_col(goose[0], configuration.columns)
# Add possible head movements for it
for px, py in get_nearest_cells(x, y):
#print ("Head possible cell on ({}, {})".format(px, py))
# If one of these head movements may lead the goose
# to eat, add tail as BODY_CELL, because it won't move.
if board[px, py] == FOOD_CELL:
x_tail, y_tail = row_col(goose[-1], configuration.columns)
#print ("Adding tail on ({}, {}) as the goose may eat".format(x_tail, y_tail))
board[x_tail, y_tail] = BODY_CELL
board[px, py] = HEAD_POSSIBLE_CELL
# Adds goose body without tail (tail is previously added only if goose may eat)
for n in goose[:-1]:
x, y = row_col(n, configuration.columns)
#print ("Body cell on ({}, {})".format(x, y))
board[x, y] = BODY_CELL
# Adding my head to the board
x, y = row_col(player_head, configuration.columns)
#print ("My head is at ({}, {})".format(x, y))
board[x, y] = MY_HEAD
# Debug board
#print (board)
# Iterate over food and geese in order to compute distances for each one
food_race = {}
for food in observation.food:
food_race[food] = {}
for i in range(4):
goose = observation.geese[i]
if len(goose) == 0:
continue
food_race[food][i] = cell_distance(goose[0], food, configuration)
# The best food is the least coveted
best_food = None
best_distance = float('inf')
best_closest_geese = float('inf')
for food in food_race:
#print ("-> Food on {}".format(row_col(food, configuration.columns)))
my_distance = food_race[food][player_index]
#print (" - My distance is {}".format(my_distance))
closest_geese = 0
for goose_id in food_race[food]:
if goose_id == player_index:
continue
if food_race[food][goose_id] <= my_distance:
closest_geese += 1
#print (" - There are {} closest geese".format(closest_geese))
if (closest_geese < best_closest_geese):
best_food = food
best_distance = my_distance
best_closest_geese = closest_geese
#print (" * This food is better")
elif (closest_geese
== best_closest_geese) and (my_distance <= best_distance):
best_food = food
best_distance = my_distance
best_closest_geese = closest_geese
#print (" * This food is better")
# Now that the best food has been found, check if the movement towards it is safe.
# Computes every available move and then check for move priorities.
if len(player_goose) > 1:
food_movements = move_towards(player_head, player_goose[1], best_food,
configuration)
else:
food_movements = move_towards(player_head, player_head, best_food,
configuration)
all_movements = get_all_movements(player_head, configuration)
# Excluding last movement reverse
food_movements = [
move for move in food_movements if move[2] != REVERSE_MOVE[last_move]
]
all_movements = [
move for move in all_movements if move[2] != REVERSE_MOVE[last_move]
]
#print ("-> Available food moves: {}".format(food_movements))
#print ("-> All moves: {}".format(all_movements))
# Trying to reach goal size of 4
if (len(player_goose) < 4):
# 1. Food movements that are safe and not closed
for food_movement in food_movements:
#print ("Food movement {}".format(food_movement))
if is_safe(food_movement,
board) and not is_closed(food_movement, board):
#print ("It's safe! Let's move {}!".format(moves[food_movement[2]]))
last_move = food_movement[2]
return moves[food_movement[2]] # Move here
# 2. Any movement safe and not closed
for movement in all_movements:
#print ("Movement {}".format(movement))
if is_safe(movement, board) and not is_closed(movement, board):
#print ("It's safe! Let's move {}!".format(moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# 3. Food movements half safe and not closed
for food_movement in food_movements:
if is_half_safe(food_movement,
board) and not is_closed(food_movement, board):
#print ("Food movement {} is half safe, I'm going {}!".format(food_movement, moves[food_movement[2]]))
last_move = food_movement[2]
return moves[food_movement[2]] # Move here
# 4. Any movement half safe and not closed
for movement in all_movements:
if is_half_safe(movement,
board) and not is_closed(movement, board):
#print ("Movement {} is half safe, I'm going {}!".format(movement, moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# 5. Food movements that are safe
for food_movement in food_movements:
#print ("Food movement {}".format(food_movement))
if is_safe(food_movement, board):
#print ("It's safe! Let's move {}!".format(moves[food_movement[2]]))
last_move = food_movement[2]
return moves[food_movement[2]] # Move here
# 6. Any movement safe
for movement in all_movements:
#print ("Movement {}".format(movement))
if is_safe(movement, board):
#print ("It's safe! Let's move {}!".format(moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# 7. Food movements half safe
for food_movement in food_movements:
if is_half_safe(food_movement, board):
#print ("Food movement {} is half safe, I'm going {}!".format(food_movement, moves[food_movement[2]]))
last_move = food_movement[2]
return moves[food_movement[2]] # Move here
# 8. Any movement half safe
for movement in all_movements:
if is_half_safe(movement, board):
#print ("Movement {} is half safe, I'm going {}!".format(movement, moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# Just trying to walk in circles
else:
# Delete food moves
food_coordinates = []
for food in food_race:
x_food, y_food = row_col(food, configuration.columns)
food_coordinates.append((x_food, y_food))
available_moves = []
for move in all_movements:
for (x_food, y_food) in food_coordinates:
if (move[0] != x_food) or (move[1] != y_food):
available_moves.append(move)
# 1. Run in circles if you can
circle_move = CIRCLE_MOVE[last_move]
for move in available_moves:
if (move[2] == circle_move) and (is_safe(
move, board)) and not (is_closed(move, board)):
last_move = move[2]
return moves[move[2]]
# 2. Any movement safe and not closed
for movement in all_movements:
#print ("Movement {}".format(movement))
if is_safe(movement, board) and not is_closed(movement, board):
#print ("It's safe! Let's move {}!".format(moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# 3. Any movement half safe and not closed
for movement in all_movements:
if is_half_safe(movement,
board) and not is_closed(movement, board):
#print ("Movement {} is half safe, I'm going {}!".format(movement, moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# 4. Any movement safe
for movement in all_movements:
#print ("Movement {}".format(movement))
if is_safe(movement, board):
#print ("It's safe! Let's move {}!".format(moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# 5. Any movement half safe
for movement in all_movements:
if is_half_safe(movement, board):
#print ("Movement {} is half safe, I'm going {}!".format(movement, moves[movement[2]]))
last_move = movement[2]
return moves[movement[2]] # Move here
# Finally, if all moves are unsafe, randomly pick one
rand_pick = np.random.randint(4) + 1
last_move = rand_pick
#print ("Yeah whatever, I'm going {}".format(moves[rand_pick]))
return moves[rand_pick]
def agent(obs_dict, config_dict):
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
player_tail = player_goose[-1]
my_X, my_Y = row_col(player_head, configuration.columns)
my_tail_X, my_tail_Y = row_col(player_tail, configuration.columns)
mask = np.zeros((configuration.rows, configuration.columns))
global last_direction
global iteration_number
print("**iteration**: {}".format(iteration_number))
print("my head at {} - {}".format([my_X, my_Y], last_direction))
iteration_number+=1
food_location = []
available_steps = {}
other_heads = [] # head of competitor geese
body_cells = []
other_tails = []
last_direction_updated = False
# add all directions
possible_positions = get_nearest_cells(my_X, my_Y)
for x, y in possible_positions:
direct = '-'
if((x-1 == my_X) or (x==0 and my_X==6)) and last_direction != 'NORTH':
direct = 'SOUTH'
elif((x+1 == my_X) or (x==6 and my_X==0)) and last_direction != 'SOUTH':
direct = 'NORTH'
elif((y-1 == my_Y) or (y==0 and my_Y==10)) and last_direction != 'WEST':
direct = 'EAST'
elif((y+1 == my_Y) or (y==10 and my_Y==0)) and last_direction != 'EAST':
direct = 'WEST'
if direct != '-':
available_steps[(x,y)] = [direct,0]
for food in observation.food:
x,y = row_col(food, configuration.columns)
mask[x, y] = 2
food_location.append([x,y])
print("food locations are {}".format(food_location))
# where other goose can come
danger_area = {}
for i in range(len(observation.geese)):
opp_goose = observation.geese[i]
if len(opp_goose) == 0:
continue
for ind, goose in enumerate(opp_goose):
x, y = row_col(goose, configuration.columns)
mask[x, y] = -1
# and mark the occupied cells if not head or tail of goose
if ind != 0 and ind != len(opp_goose) - 1:
body_cells.append([x,y])
if ind != 0 and ind == len(opp_goose) - 1:
other_tails.append([x,y])
if (x, y) in available_steps.keys():
# let's remove all cells that are not available from available steps
available_steps.pop((x,y))
if (x, y) in danger_area.keys():
# let's remove all cells that are not available from danger steps
danger_area.pop((x,y))
if not len(available_steps) and not len(danger_area): #if steps is empty kill by NORTH direction
print("kill")
if player_head!=player_tail and (my_tail_X, my_tail_Y) in possible_positions:
last_direction = get_direction((my_X, my_Y), (my_tail_X, my_tail_Y))
else:
last_direction = 'NORTH'
last_direction_updated = True
break
if last_direction_updated:
break
if i != player_index:
x, y = row_col(opp_goose[0], configuration.columns)
# add competition goose head
other_heads.append([x, y])
possible_moves = get_nearest_cells(x, y) # head can move anywhere
for [x, y] in possible_moves:
if (x,y) in available_steps.keys() and [x,y] in food_location:
danger_area[(x,y)] = available_steps[(x,y)]
available_steps.pop((x, y)) #only safe moves
food_location.remove([x,y]) #don't consider this food
elif (x,y) in available_steps.keys() and len(available_steps)+len(danger_area) > 1:
danger_area[(x,y)] = available_steps[(x,y)]
available_steps.pop((x, y)) #only safe moves
if last_direction_updated:
break
print("body cells are {}".format(body_cells))
print("available steps are {}".format(available_steps))
print("other heads at {}".format(other_heads))
print("other tails at {}".format(other_tails))
print("danger area at {}".format(danger_area))
my_head = [my_X, my_Y]
if not last_direction_updated:
last_direction = my_move(food_location, available_steps, other_heads, my_head, last_direction, body_cells, danger_area, mask, other_tails)
last_direction_updated = True
return last_direction
def encodeObservations(self, obs_dict, config_dict):
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
return self._state.local(observation, configuration)
def agent(obs_dict, config_dict):
global which_turn
which_turn += 1
t0= time.clock()
f = open("./myfile1.txt", "a")
f.write("\n")
f.write("Starting file" + "\n")
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
#player_goose is list of all bodypart positions
player_head = player_goose[0]
player_row, player_column = row_col(player_head, configuration.columns)
start = (player_row, player_column)
f.write("Commencing turn: " + str(which_turn))
with open('./myfile2.txt', 'a') as testfile:
testfile.write("Player_index: " + str(player_index) + " [0: White, 1: Blue, 2: Green, 3: Red], " + "\n")
f.write("Player_index: " + str(player_index) + " [0: White, 1: Blue, 2: Green, 3: Red], " + "\n")
"""This agent always moves toward observation.food[0] but does not take advantage of board wrapping"""
Matrix, MatrixNoFood, MatrixHeadAvoid1, MatrixHeadAvoid2 = fill_matrix(observation, configuration)
f.write("Matrix filled" + "\n")
[MatrixNoFood, MatrixHeadAvoid1, MatrixHeadAvoid2], start, shift = shift_matrix([MatrixNoFood, MatrixHeadAvoid1, MatrixHeadAvoid2], start)
f.write("Matrix shifted" + "\n")
f.write("Start coords " + str(start) + "\n")
#MatrixToUse = MatrixNoFood
#MatrixToUse = MatrixHeadAvoid2
#MatrixToUse = MatrixHeadAvoid1
#MatrixBackup1 = MatrixHeadAvoid1
#MatrixBackup2 = MatrixNoFood
#MatrixsToUse = [MatrixHeadAvoid2, MatrixHeadAvoid1, MatrixNoFood]
MatrixsToUse = [MatrixHeadAvoid1, MatrixNoFood]
#will replace player_row, player_column with 2d array that has all positions filled with all player information
path = path_to_closest_food(observation, configuration, MatrixsToUse, start, shift)
if path == "no path":
player_index = observation.index
player_goose = observation.geese[player_index]
player_end = player_goose[len(player_goose)-1]
player_end_row, player_end_column = row_col(player_end, configuration.columns)
for MatrixToUse in MatrixsToUse:
if path == "no path":
if start == (player_end_row, player_end_column):
path = astar(MatrixToUse, start, (player_end_row+3, player_end_column+5))
else:
path = astar(MatrixToUse, start, (player_end_row, player_end_column))
f.write("Path: " + str(path) + "\n")
whichMove = which_direction(path)
t1 = time.clock() - t0
f.write("This is turn: " + str(int(which_turn)) + " , chosen movement: " )
f.write(str(whichMove))
f.write(" which took " + str(t1) + " seconds")
f.close()
return whichMove
def agent(obs_dict, config_dict):
"""This agent always moves toward observation.food[0] but does not take advantage of board wrapping"""
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
maze_col = configuration.columns
maze_rows = configuration.rows
maze = np.zeros([maze_rows, maze_col])
print("Player")
# Define my player
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
player_row, player_column = row_col(player_head, configuration.columns)
start = (player_row, player_column)
print(start)
#Define Food
hipotenuse_low = 14
closest_food = observation.food[0]
for food in observation.food:
food_row, food_column = row_col(food, configuration.columns)
hipotenuse = sqrt((player_row - food_row)**2 +
(player_column - food_column)**2)
if hipotenuse < hipotenuse_low:
hipotenuse_low = hipotenuse
closest_food = food
end = row_col(closest_food, configuration.columns)
# Define the enemies players
enemies_indexes = observation.geese
del enemies_indexes[player_index]
for enemies in enemies_indexes:
for enemies_pos in enemies:
enemie_row, enemie_col = row_col(enemies_pos,
configuration.columns)
maze[enemie_row, enemie_col] = 1
for player_pos in player_goose[1:]:
player_rw, player_col = row_col(player_pos, configuration.columns)
maze[player_rw, player_col] = 1
path = astar(maze, start, end, False)
next_move = path[1]
print(next_move)
print(player_column)
print(player_row)
x = int(next_move[1]) - int(player_column)
y = int(next_move[0]) - int(player_row)
print(x, y)
print(maze)
if x == -1 and y == 0:
print("WEST")
return Action.WEST.name
if x == 1 and y == 0:
print("EAST")
return Action.EAST.name
if x == 0 and y == -1:
print("NORTH")
return Action.NORTH.name
if x == 0 and y == 1:
print("SOUTH")
return Action.SOUTH.name
def agent(obs_dict, config_dict, gindex):
#################################################
# State retrieval
#################################################
#print("-----------")
global last_action
conf = Configuration(config_dict)
obs = Observation(obs_dict)
step = obs.step + 1
my_idx = gindex
my_goose = obs.geese[my_idx]
my_head = my_goose[0]
my_row, my_col = row_col(position=my_head, columns=conf.columns)
if DEBUG:
print("---------- Step #" + str(step), "- Player #" + str(obs.index))
#################################################
# Map update
#################################################
board = np.zeros((7, 11), dtype=int)
# Add food to board
for food in obs.food:
food_row, food_col = row_col(position=food, columns=conf.columns)
board[food_row, food_col] += FOOD_REWARD
'''if DEBUG:
print("food", food_row, food_col)'''
# Iterate over geese to add geese data to board
nb_geese = len(obs.geese)
geese_lengths = []
for i in range(nb_geese):
'''if DEBUG:
print("--- Goose #" + str(i))'''
goose = obs.geese[i]
potential_food_head = None
# Iterate over cells of current goose
goose_len = len(goose)
geese_lengths.append(goose_len)
'''if DEBUG:
print("--- Goose #" + str(i) + " len " + str(goose_len))'''
for j in range(goose_len):
'''if DEBUG:
print("--- Goose #" + str(i) + " cell " + str(j))'''
goose_cell = goose[j]
goose_row, goose_col = row_col(position=goose_cell,
columns=conf.columns)
# Check for food on neighbour cells when handling head
if j == 0:
potential_heads = get_neighbours(goose_row, goose_col)
for potential_head in potential_heads:
for food in obs.food:
food_row, food_col = row_col(position=food,
columns=conf.columns)
if potential_head == (food_row, food_col):
potential_food_head = potential_head
# Update rewards linked to body/tail
if j < goose_len - 1:
# Body or head
board[goose_row, goose_col] += BODY_REWARD
'''if DEBUG:
print("--- Goose #" + str(i) + " cell " + str(j) + " add BODY_REWARD")'''
else:
# Tail : may not move if goose eats
if potential_food_head is not None:
board[goose_row, goose_col] += TAIL_REWARD
'''if DEBUG:
print("--- Goose #" + str(i) + " cell " + str(j) + " add TAIL_REWARD")'''
# Update potential villain head positions
if (i != my_idx) & (goose_len > 0):
if potential_food_head is not None:
# Head will prolly go to the food
for potential_head in potential_heads:
if potential_head == potential_food_head:
if (board[potential_head[0], potential_head[1]] != BODY_REWARD) & \
(board[potential_head[0], potential_head[1]] != TAIL_REWARD):
board[
potential_head[0],
potential_head[1]] += PROBABLE_HEAD_FOOD_REWARD
'''if DEBUG:
print("--- Goose #" + str(i) + " cell " + str(j) + " add PROBABLE_HEAD_FOOD_REWARD")'''
else:
if (board[potential_head[0], potential_head[1]] != BODY_REWARD) & \
(board[potential_head[0], potential_head[1]] != TAIL_REWARD):
board[potential_head[0], potential_head[
1]] += IMPROBABLE_HEAD_FOOD_REWARD
'''if DEBUG:
print("--- Goose #" + str(i) + " cell " + str(j) + " add IMPROBABLE_HEAD_FOOD_REWARD")'''
else:
# Standard potential head reward
for potential_head in potential_heads:
if (board[potential_head[0], potential_head[1]] != BODY_REWARD) & \
(board[potential_head[0], potential_head[1]] != TAIL_REWARD):
board[potential_head[0],
potential_head[1]] += POTENTIAL_HEAD_STD_REWARD
'''if DEBUG:
print("--- Goose #" + str(i) + " cell " + str(j) + " add POTENTIAL_HEAD_STD_REWARD")'''
# Check if I'm the current longest Goose
if (len(my_goose) >= max(geese_lengths) - 3) & (step > 8):
# Chasing my tail as a defensive action makes sense
my_tail_row, my_tail_col = row_col(position=my_goose[-1],
columns=conf.columns)
board[my_tail_row, my_tail_col] += TAIL_CHASE_REWARD
'''if DEBUG:
print("Adding TAIL_CHASE_REWARD for me")'''
# Diffuse values in adjacent cells
if DEBUG:
print("Initial board :")
print(board)
new_board = board.copy()
for i in range(7):
for j in range(11):
value = board[i, j]
if value > DIFFUSE_START:
# Should diffuse positive value
neighbours = get_neighbours(i, j)
for neighbour in neighbours:
# Level 1
new_board[neighbour] += (2 * DIFFUSE_POS_REWARD)
# Level 2
neighbours_lvl2 = get_neighbours(neighbour[0],
neighbour[1])
for neighbour_lvl2 in neighbours_lvl2:
new_board[neighbour_lvl2] += DIFFUSE_POS_REWARD
elif value < -DIFFUSE_START:
# Should diffuse negative value
neighbours = get_neighbours(i, j)
for neighbour in neighbours:
# Level 1
new_board[neighbour] += (2 * DIFFUSE_NEG_REWARD)
# Level 2
neighbours_lvl2 = get_neighbours(neighbour[0],
neighbour[1])
for neighbour_lvl2 in neighbours_lvl2:
new_board[neighbour_lvl2] += DIFFUSE_NEG_REWARD
board = new_board
# Add last_action data to board
if last_action is not None:
if last_action == Action.SOUTH.name:
board[(my_row + 6) % 7, my_col] += REVERSE_LAST_REWARD
elif last_action == Action.NORTH.name:
board[(my_row + 8) % 7, my_col] += REVERSE_LAST_REWARD
elif last_action == Action.EAST.name:
board[my_row, (my_col + 10) % 11] += REVERSE_LAST_REWARD
elif last_action == Action.WEST.name:
board[my_row, (my_col + 12) % 11] += REVERSE_LAST_REWARD
'''if DEBUG:
print("Adding REVERSE_LAST_REWARD for me")'''
if DEBUG:
print("Final board :")
print(board)
#################################################
# Choose best action
#################################################
chosen_action = None
rewards = []
potential_next = get_neighbours(my_row, my_col)
for cell in potential_next:
rewards.append(board[cell])
choice = np.argmax(rewards)
if choice == 0:
chosen_action = Action.NORTH.name
elif choice == 1:
chosen_action = Action.WEST.name
elif choice == 2:
chosen_action = Action.SOUTH.name
else:
chosen_action = Action.EAST.name
if DEBUG:
print("chosen_action", chosen_action)
last_action = chosen_action
return chosen_action
def agent(obs_dict, config_dict):
"""This agent always moves toward observation.food[0] but does not take advantage of board wrapping"""
global last_move
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
player_row, player_column = row_col(player_head, configuration.columns)
possible_moves = [
Action.SOUTH.name, Action.NORTH.name, Action.WEST.name,
Action.EAST.name
]
possible_move = None
oldFoodDistanceToMe = 9999
def getFoodsIndex():
foods = []
for i in range(len(observation.food)):
food = observation.food[i]
foods.append(row_col(food, configuration.columns))
return foods
def getGoose(index):
goose = []
for i in range(len(observation.geese[index])):
goose.append(
row_col(observation.geese[index][i], configuration.columns))
return goose
def getGeesesIndex():
geeses = []
for geese in observation.geese:
geeses.append(row_col(geese[0], configuration.columns))
return geeses
def randomMove(moves):
if len(moves) > 1:
return moves[randint(0, len(moves) - 1)]
return moves[0]
def countMoves(x1, y1, x2, y2):
return abs((x1 - x2) + (y1 - y2))
def willCollide(x, y, action):
#print(f'Goose[{player_index} in player_row: {player_row}, player_column: {player_column} move to {action} and can collide with x: {x}, y: {y}')
if action == Action.WEST.name:
if player_column - 1 == y and x == player_row:
return True
if player_column == 0 and y == 10 and x == player_row:
return True
if action == Action.EAST.name:
if player_column + 1 == y and x == player_row:
return True
if player_column == 10 and y == 0 and x == player_row:
return True
if action == Action.SOUTH.name:
if player_row + 1 == x and y == player_column:
return True
if player_row == 6 and x == 0 and y == player_column:
return True
if action == Action.NORTH.name:
if player_row - 1 == x and y == player_column:
return True
if player_row == 0 and x == 6 and y == player_column:
return True
return False
def opposite(action):
if action == Action.NORTH.name:
return Action.SOUTH.name
if action == Action.SOUTH.name:
return Action.NORTH.name
if action == Action.EAST.name:
return Action.WEST.name
if action == Action.WEST.name:
return Action.EAST.name
possible_moves = [
Action.SOUTH.name, Action.NORTH.name, Action.WEST.name,
Action.EAST.name
]
if last_move[player_index] is not None:
print(
f'Geese {player_index} removes opposite last move: {opposite(last_move[player_index])}'
)
possible_moves.remove(opposite(last_move[player_index]))
for geese_row, geese_column in getGeesesIndex():
for food_row, food_column in getFoodsIndex():
foodDistanceToMe = countMoves(food_row, food_column, player_row,
player_column)
if oldFoodDistanceToMe > foodDistanceToMe:
oldFoodDistanceToMe = foodDistanceToMe
else:
continue
geeseDistanceToFood = countMoves(geese_row, geese_column, food_row,
food_column)
if foodDistanceToMe < geeseDistanceToFood:
if food_row > player_row:
possible_move = Action.SOUTH.name
if food_row < player_row:
possible_move = Action.NORTH.name
if food_column > player_column:
possible_move = Action.EAST.name
if food_column < player_column:
possible_move = Action.WEST.name
print(
f'Geese {player_index} based by food, possible move: {possible_move}'
)
if possible_move not in possible_moves:
print(
f'Geese {player_index}, possible move: {possible_move} is banned.')
possible_move = randomMove(possible_moves)
for i in range(len(observation.geese)):
geese_index = getGoose(i)
j = 0
while j < len(geese_index):
collision = False
x, y = geese_index[j]
if i != player_index:
print(f'geese {player_index} -> geese {i} part {j}')
#print(f'goose[{player_index}] in {getGoose(player_index)[0]} with possible move: {possible_move} can colide with goose[{i}] in x: {x} and y: {y}')
while willCollide(x, y, possible_move):
collision = True
print(
f'goose[{player_index}] in {getGoose(player_index)[0]} with possible move: {possible_move} will colide with goose[{i}] in x: {x} and y: {y}'
)
possible_moves.remove(possible_move)
possible_move = randomMove(possible_moves)
print(f'now goose[{player_index}] will to {possible_move}')
j = 0
if not collision:
j += 1
for x, y in getGoose(player_index):
while willCollide(x, y, possible_move):
print(f'BODY HIT!!!')
possible_moves.remove(possible_move)
possible_move = randomMove(possible_moves)
if len(possible_moves) == 1:
print(
f'Geese {player_index} just remain this move: {possible_moves[0]}')
possible_move = possible_moves[0]
last_move[player_index] = possible_move
print(
f'Geese {player_index}: {getGoose(player_index)} move to {possible_move}'
)
return possible_move
def agent(obs_dict, config_dict):
global ralph_last_action
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
#print("Player")
# Define my player
player_index = observation.index
my_goose = Goose(observation.geese[player_index], configuration)
player_row, player_column = my_goose.get_head_pos()
#print("Enemies")
# Define the enemies players
enemies_indexes = []
for i in range(0, len(observation.geese)):
if i != player_index:
enemies_indexes.append(i)
enemies_geese = []
for enemy in enemies_indexes:
enemies_geese.append(Goose(observation.geese[enemy], configuration))
#print("Food")
# Get the nearest food position
food_row, food_column = get_min_distance_food_pos(my_goose.get_head(),
observation.food,
configuration.columns)
#print("Possible Moves")
# List the possible moves
possible_moves = ['SOUTH', 'NORTH', 'WEST', 'EAST']
if (ralph_last_action[player_index] in possible_moves):
possible_moves.remove(ralph_last_action[player_index])
#print("Possible Moves and colisions")
moves = []
for i in range(0, len(possible_moves)):
move = possible_moves[i]
n_r, n_c = my_goose.get_next_position(move)
if (not is_any_enemy_there(enemies_geese, n_r, n_c)
and not my_goose.am_i_there(n_r, n_c)
and not will_be_any_enemy_there(enemies_geese, n_r, n_c)):
moves.append(move)
#print("Possible Moves by food")
# From the remaining possible moves, wich one get me nearest to the food?
min_food_move = 0
if (len(moves) > 0):
for i in range(0, len(moves)):
n_r1, n_c1 = my_goose.get_next_position(moves[i])
n_r2, n_c2 = my_goose.get_next_position(moves[min_food_move])
if (get_distance(food_row, food_column, n_r1, n_c1) < get_distance(
food_row, food_column, n_r2, n_c2)):
min_food_move = i
#print("Select Move")
if (len(moves) > 0 and min_food_move is not None):
return_action = moves[min_food_move]
else:
return_action = 'NORTH'
#print("Last Move")
# Refresh the last move
ralph_last_action[player_index] = opposite(return_action)
#print(return_action)
#print()
return return_action
def agent(obs_dict, config_dict):
global last_step
# State retrieval
observation = Observation(obs_dict)
configuration = Configuration(config_dict)
player_index = observation.index
player_goose = observation.geese[player_index]
player_head = player_goose[0]
player_row, player_column = row_col(player_head, configuration.columns)
# Map creation
# 0 - empty cells
# -1 - obstacles
# -4 - possible obstacles
# -2 - food
# -3 - head
# 1,2,3,4 - reachable on the current step cell, number is the id of the first step direction
table = np.zeros((7, 11))
legend = {1: 'SOUTH', 2: 'NORTH', 3: 'EAST', 4: 'WEST'}
# let's add food to the map
for food in observation.food:
x, y = row_col(food, configuration.columns)
table[x, y] = -2 # food
# let's add all cells that are forbidden
for i in range(4):
opp_goose = observation.geese[i]
if len(opp_goose) == 0:
continue
is_close_to_food = False
if i != player_index:
x, y = row_col(opp_goose[0], configuration.columns)
possible_moves = get_nearest_cells(x, y) # head can move anywhere
for x, y in possible_moves:
if table[x, y] == -2:
is_close_to_food = True
table[
x,
y] = -4 # Cells where opponent might possibly go next step
# usually we ignore the last tail cell but there are exceptions
tail_change = -1
if obs_dict['step'] % 40 == 39:
tail_change -= 1
# we assume that the goose will eat the food
if is_close_to_food:
tail_change += 1
if tail_change >= 0:
tail_change = None
for n in opp_goose[:tail_change]:
x, y = row_col(n, configuration.columns)
table[x, y] = -1 # forbidden cells
# going back is forbidden according to the new rules
x, y = row_col(player_head, configuration.columns)
if last_step is not None:
if last_step == 1:
table[(x + 6) % 7, y] = -1
elif last_step == 2:
table[(x + 8) % 7, y] = -1
elif last_step == 3:
table[x, (y + 10) % 11] = -1
elif last_step == 4:
table[x, (y + 12) % 11] = -1
# add head position
table[x, y] = -3
# the first step toward the nearest food
step = int(find_closest_food(table))
# if there is not available steps try to go to possibly dangerous cell
if step not in [1, 2, 3, 4]:
x, y = row_col(player_head, configuration.columns)
if table[(x + 8) % 7, y] == -4:
step = 1
elif table[(x + 6) % 7, y] == -4:
step = 2
elif table[x, (y + 12) % 11] == -4:
step = 3
elif table[x, (y + 10) % 11] == -4:
step = 4
# else - do a random step and lose
else:
step = np.random.randint(4) + 1
last_step = step
return legend[step]
def override_interpreter(state, env):
configuration = Configuration(env.configuration)
columns = configuration.columns
rows = configuration.rows
min_food = configuration.min_food
state[0].observation = shared_observation = Observation(state[0].observation)
hits = 0
# Reset the environment.
if env.done:
agent_count = len(state)
heads = sample(range(columns * rows), agent_count)
shared_observation["geese"] = [[head] for head in heads]
food_candidates = set(range(columns * rows)).difference(heads)
# Ensure we only place as many food as there are open squares
min_food = min(min_food, len(food_candidates))
shared_observation["food"] = sample(food_candidates, min_food)
return state
geese = shared_observation.geese
food = shared_observation.food
# If there is no last state, reuse current state so that current action is never the opposite of the last action.
last_state = env.steps[-1] if len(env.steps) > 1 else state
# Apply the actions from active agents.
for index, agent in enumerate(state):
if agent.status != "ACTIVE":
if agent.status != "INACTIVE" and agent.status != "DONE":
# ERROR, INVALID, or TIMEOUT, remove the goose.
geese[index] = []
continue
action = Action[agent.action]
# Check action direction
last_agent = last_state[index]
last_action = Action[last_agent["action"]] if "action" in last_agent else action
if last_action == action.opposite():
env.debug_print(f"Opposite action: {agent.observation.index, action, last_action}")
agent.status = "DONE"
geese[index] = []
hits +=1
continue
goose = geese[index]
head = translate(goose[0], action, columns, rows)
# Consume food or drop a tail piece.
if head in food:
food.remove(head)
else:
goose.pop()
# Self collision.
if head in goose:
env.debug_print(f"Body Hit: {agent.observation.index, action, head, goose}")
agent.status = "DONE"
geese[index] = []
hits +=1
continue
while len(goose) >= configuration.max_length:
# Free a spot for the new head if needed
goose.pop()
# Add New Head to the Goose.
goose.insert(0, head)
# If hunger strikes remove from the tail.
if len(env.steps) % configuration.hunger_rate == 0:
if len(goose) > 0:
goose.pop()
if len(goose) == 0:
env.debug_print(f"Goose Starved: {action}")
agent.status = "DONE"
hits +=1
continue
goose_positions = histogram(
position
for goose in geese
for position in goose
)
# Check for collisions.
for index, agent in enumerate(state):
goose = geese[index]
if len(goose) > 0:
head = geese[index][0]
if goose_positions[head] > 1:
env.debug_print(f"Goose Collision: {agent.action}")
agent.status = "DONE"
geese[index] = []
hits +=1
# Add food if min_food threshold reached.
needed_food = min_food - len(food)
if needed_food > 0:
collisions = {
position
for goose in geese
for position in goose
}
available_positions = set(range(rows * columns)).difference(collisions).difference(food)
# Ensure we don't sample more food than available positions.
needed_food = min(needed_food, len(available_positions))
food.extend(sample(available_positions, needed_food))
# Set rewards after deleting all geese to ensure that geese don't receive a reward on the turn they perish.
for index, agent in enumerate(state):
if agent.status == "ACTIVE":
agent.reward = (configuration.max_length + 1)*(hits+1) + len(geese[index]) - sum((len(goose) for goose in geese))
# Adding 1 to len(env.steps) ensures that if an agent gets reward 4507, it died on turn 45 with length 7.
# If only one ACTIVE agent left, set it to DONE.
active_agents = [a for a in state if a.status == "ACTIVE"]
if len(active_agents) == 1:
agent = active_agents[0]
agent.status = "DONE"
return state