def get_disallowed_actions(self, obs):
disallowed_actions = []
refinery_y, refinery_x = get_refineries(self.unit_type)
geyser_top_left = geysers[0]
if geyser_top_left[0] in refinery_x and geyser_top_left[1] in refinery_y:
disallowed_actions.append(get_action_id(ACTION_BUILD_REFINERY_TOP_LEFT))
geyser_top_right = geysers[1]
if geyser_top_right[0] in refinery_x and geyser_top_right[1] in refinery_y:
disallowed_actions.append(get_action_id(ACTION_BUILD_REFINERY_TOP_RIGHT))
geyser_bottom_left = geysers[2]
if geyser_bottom_left[0] in refinery_x and geyser_bottom_left[1] in refinery_y:
disallowed_actions.append(get_action_id(ACTION_BUILD_REFINERY_BOTTOM_LEFT))
geyser_bottom_right = geysers[3]
if geyser_bottom_right[0] in refinery_x and geyser_bottom_right[1] in refinery_y:
disallowed_actions.append(get_action_id(ACTION_BUILD_REFINERY_BOTTOM_RIGHT))
if self.supply_depots >= len(depots):
disallowed_actions.append(get_action_id(ACTION_BUILD_SUPPLY_DEPOT))
if p.get_food_cap(obs) == p.get_food_used(obs):
disallowed_actions.append(get_action_id(ACTION_BUILD_SCV_START_CC))
disallowed_actions.append(get_action_id(ACTION_BUILD_SCV_NEW_CC))
if get_command_center_amount(self.unit_type) == 2:
disallowed_actions.append(get_action_id(ACTION_BUILD_CC))
return disallowed_actions
def handle_last_action(self, obs):
cm = str(p.get_minerals(obs))
cv = str(p.get_vespene(obs))
iwc = str(p.get_idle_worker_count(obs))
depot_count = str(self.supply_depots)
refinery_count = str(self.refineries)
command_centers = str(get_command_center_amount(self.unit_type))
food_used = str(p.get_food_used(obs))
food_cap = str(p.get_food_cap(obs))
sp_attempts = str(self.build_supply_depot_attempts)
sp_attempts_f = str(self.build_supply_depot_attempts_failed)
refinery_attempts = str(self.build_refinery_attempts)
refinery_attempts_f = str(self.build_refinery_attempts_failed)
cc_attempts = str(self.build_cc_attempts)
cc_attempts_f = str(self.build_cc_attempt_failed)
score = cs.get_score(obs)
data = [sp_attempts, sp_attempts_f, depot_count, refinery_attempts, refinery_attempts_f, refinery_count, cc_attempts, cc_attempts_f, cm, cv, iwc, command_centers, food_used + "/" + food_cap, str(score)]
with open('/home/kenn/Development/sc2-bot/CustomAgents/scores.txt', 'a+') as f:
f.write('{0[0]:<15}{0[1]:<15}{0[2]:<15}{0[3]:<15}{0[4]:<15}{0[5]:<15}{0[6]:<15}{0[7]:<15}{0[8]:<15}{0[9]:<15}{0[10]:<15}{0[11]:<15}{0[12]:<15}{0[13]:<15}\n'.format(data))
# If we score less than 4000 we are doing so poorly, we want to learn that it was very bad.
# Symbolizes the ultimate loss.
self.qlearn.learn(str(self.previous_state), self.previous_action, int(score) - 4000, 'terminal')
self.previous_state = None
self.previous_action = None
self.move_number = 0
self.supply_depots = 0
self.refineries = 0
self.builder_iterator = 0
self.build_supply_depot_attempts = 0
self.build_supply_depot_attempts_failed = 0
self.build_refinery_attempts = 0
self.build_refinery_attempts_failed = 0
self.build_cc_attempts = 0
self.build_cc_attempt_failed = 0
self.initializing = 0
self.qlearn.q_table.to_pickle(DATA_FILE + '.gz', 'gzip')
return actions.FunctionCall(_NOOP, [])
def step(self, obs):
super(CollectMineralsAndGas, self).step(obs)
self.unit_type = obs.observation['screen'][_UNIT_TYPE]
if obs.last():
return self.handle_last_action(obs)
if obs.first():
return self.handle_first_action(obs)
if self.initializing < 2:
return self.handle_initial_action(obs)
if self.move_number == 0:
self.update_buildings_built()
supply_used = p.get_food_used(obs)
supply_cap = p.get_food_cap(obs)
scvs_left = get_approx_scvs_in_rectangle_count(self.unit_type, (0, 0), (41, 83))
scvs_right = get_approx_scvs_in_rectangle_count(self.unit_type, (42, 0), (83, 83))
current_state = np.zeros(6)
# Available supply min to reduce state space
current_state[0] = min(supply_cap - supply_used, 7)
# Detect how many times we have 100 minerals up to maximum 4 times (Afford a CC) to reduce our state space
current_state[1] = min(int(p.get_minerals(obs) / 100.0), 4)
# Any idle workers min to reduce state space
current_state[2] = min(p.get_idle_worker_count(obs), 3)
# scvs left side to scvs right side difference
scv_side_diff = scvs_left - scvs_right
current_state[3] = max(-7, scv_side_diff) if scv_side_diff < 0 else min(8, scv_side_diff)
# Number of CCs
current_state[4] = get_command_center_amount(self.unit_type)
# Number of supply depots
current_state[5] = get_supply_depot_amount(self.unit_type)
if self.previous_action is not None:
# reward = self.get_reward(obs)
self.qlearn.learn(str(self.previous_state), self.previous_action, 0, str(current_state))
disallowed_actions = self.get_disallowed_actions(obs)
rl_action = self.qlearn.choose_action(str(current_state), disallowed_actions)
action_name = get_smart_action(int(rl_action))
action_function, actual_action = self.apply_action(obs, action_name)
action_id = rl_action
if actual_action is None:
action_function = actions.FunctionCall(_NOOP, [])
action_id = get_action_id(ACTION_DO_NOTHING)
self.previous_state = current_state
self.previous_action = action_id
self.move_number += 1
return action_function
elif self.move_number == 1:
action_name = get_smart_action(self.previous_action)
action_function, adjusted_action = self.apply_action(obs, action_name)
if adjusted_action is None:
action_function = actions.FunctionCall(_NOOP, [])
self.previous_action = get_action_id(ACTION_DO_NOTHING)
self.move_number += 1
return action_function
elif self.move_number == 2:
action_name = get_smart_action(self.previous_action)
action_function, adjusted_action = self.apply_action(obs, action_name)
if adjusted_action is None:
action_function = actions.FunctionCall(_NOOP, [])
self.previous_action = get_action_id(ACTION_DO_NOTHING)
self.move_number = 0
return action_function
return actions.FunctionCall(_NOOP, [])