def step(self, action, debug=0):
if type(action) is np.ndarray:
action = np.argmax(action)
is_jump = action == 1
move = self.innerAI.select_move(self.observation,
is_jump=is_jump,
debug=debug)
if move[0] == 'skip':
return self.observation, -10, True, {'nothing': 'there'}
self.observation, mv_error = apply_move(move, AGENT_ID,
self.observation)
if debug:
print(self.state)
#Greedy AI processing
move = self.enemyAI.select_move(self.observation, debug)
self.observation, _ = apply_move(move, OPPENENT_ID, self.observation)
done = is_done(self.observation)
self.reward = winner(AGENT_ID, self.observation)
if (mv_error == True):
done = True
self.reward = -20
if debug:
print(done)
# self.action_space = moves(1, self.state) - THIS SHOULD BE STATIC I BELIEVE
return self.observation, self.reward, done, {'nothing': 'there'}
def step(self, action, debug=0):
if debug:
print(self.state)
move = self.action_to_move(action)
self.observation, mv_error = apply_move(move, AGENT_ID,
self.observation)
if debug:
print(self.state)
#Greedy AI processing
move = self.enemyAI.select_move(self.observation, debug)
self.observation, _ = apply_move(move, OPPENENT_ID, self.observation)
reward = score(AGENT_ID, self.observation)
done = is_done(self.observation)
if (mv_error == True):
done = True
reward = WRONG_MOVE_PUNISHMENT
if debug:
print(done)
# self.action_space = moves(1, self.state) - THIS SHOULD BE STATIC I BELIEVE
return self.observation, reward, done, 0
def minimax(self, observation, depth=-1):
player = env.current_player(observation)
if env.is_done(observation) or depth == 0:
return env.score(self.player, observation)
else:
actions = env.legal_moves(player, observation)
scores = []
for a in actions:
self.states_evaluated += 1
s, _ = env.apply_move(a, observation)
score = self.minimax(s, depth=depth - 1)
scores.append(score)
if player == self.player:
maximum = np.argmax(scores)
return scores[maximum]
else:
minimum = np.argmin(scores)
return scores[minimum]
def look_forward(self, observation):
all_my_moves = env.legal_moves(self.player, observation)
actions = []
for m in all_my_moves:
self.states_evaluated += 1
new_observation, _ = env.apply_move(m, self.player, observation)
score = self.minimax(new_observation, self.ply - 1)
actions.append({
'action':
m,
'score':
score,
'immediate_score':
env.score(self.player, new_observation)
})
best_action = max(actions,
key=lambda p: (p['score'], p['immediate_score']))
return best_action['action']