def choose_act(self, mode, info=False):
current_state_as_dict = StateAsDict(self.env.current_state_of_the_game)
list_of_actions = self.env.action_space.list_of_actions
if list_of_actions:
best_action = None
best_action_value = -float('inf')
for action in list_of_actions:
state_copy = current_state_as_dict.to_state()
action.execute(state_copy)
state_copy.change_active_player()
# print('*******************')
current_value = self.evaluator.evaluate(state_copy)
# print(f'State_copy = {StateAsDict(state_copy)}')
# print(f'Action = {action} val = {current_value}')
# print('------------------------------------')
if current_value > best_action_value:
best_action_value = current_value
best_action = action
if not info:
return best_action
if info:
return best_action, best_action_value
else:
if not info:
return None
if info:
return None, -1
def choose_act(self, mode, info=False):
current_state_as_dict = StateAsDict(self.env.current_state_of_the_game)
list_of_actions = self.env.action_space.list_of_actions
if list_of_actions:
best_action = None
best_action_value = -100
for action in list_of_actions:
state_copy = current_state_as_dict.to_state()
action.execute(state_copy)
current_value = self.model.get_value(state_copy)
if current_value > best_action_value:
best_action_value = current_value
best_action = action
if not info:
return best_action
if info:
return best_action, best_action_value
else:
if not info:
return None
if info:
return None, -1
class DeterministicObservation(SplendorObservation):
def __init__(self, state: State):
super().__init__('deterministic')
self.observation_dict = StateAsDict(state)
def recreate_state(self):
return self.observation_dict.to_state()
def generate_all_tree_data_as_list(self,
confidence_threshold: float = 0.1,
count_threshold: int = 6,
confidence_limit: int = 2):
self.clean_memory()
# BFS
kiu = [self.root]
confidence_count = 0
print('Collecting tree data.')
X = []
Y = []
while len(kiu) > 0:
# take first:
node_to_eval = kiu.pop(0)
if node_to_eval.value_acc.count() > 0:
for child in node_to_eval.children:
# if child.value_acc.count() >= count_threshold or child.value_acc.perfect_value is not None:
if child.value_acc.get_confidence(
) >= confidence_threshold:
confidence_count += 1
if (
child.value_acc.get_confidence() >= confidence_threshold and confidence_count <= confidence_limit) \
or child.value_acc.count() >= count_threshold:
kiu.append(child)
child_state_as_dict = StateAsDict(child.return_state())
current_state = child_state_as_dict.to_state()
current_state_inversed = child_state_as_dict.to_state()
current_state_inversed.change_active_player()
current_value = child.value_acc.get()
X.append(current_state)
Y.append(current_value)
X.append(current_state_inversed)
Y.append(-current_value)
return {'state': X, 'mcts_value': Y}
def generate_all_tree_data(self):
self.clean_memory()
all_code = ''
# BFS
kiu = [self.root]
print('Collecting tree data.')
while len(kiu) > 0:
# take first:
node_to_eval = kiu.pop(0)
if node_to_eval.value_acc.count() > 0:
for child in node_to_eval.children:
if child.value_acc.count() > 0:
kiu.append(child)
child_state_as_dict = StateAsDict(child.return_state())
self.stats_dataframe = self.stats_dataframe.append(
{
'state': child_state_as_dict.to_state(),
'mcts_value': child.value_acc.get()
},
ignore_index=True)
return self.stats_dataframe
def load_state_from_dict(self, state_as_dict: StateAsDict):
self.current_state_of_the_game = state_as_dict.to_state()
self.is_done = False
'name': 'RandomAgent - uniform_on_types '
},
'board': {
'nobles_on_board': {104, 108, 102},
'cards_on_board': {33, 4, 69, 71, 72, 41, 9, 10, 44, 79, 86, 26},
'gems_on_board': [2, 4, 3, 3, 4, 3],
'deck_order': [{
'Row.CHEAP': [
38, 0, 76, 18, 55, 19, 3, 40, 7, 43, 23, 39, 37, 42, 77, 75,
25, 59, 2, 5, 54, 73, 56, 21, 57, 22, 1, 36, 6, 60, 61, 78, 74,
58, 24
]
}, {
'Row.MEDIUM': [
12, 49, 47, 31, 45, 62, 11, 81, 46, 80, 66, 67, 48, 83, 82, 65,
8, 84, 13, 29, 27, 64, 85, 63, 30
]
}, {
'Row.EXPENSIVE':
[35, 53, 87, 88, 15, 14, 50, 17, 16, 68, 70, 32, 52, 51, 89, 34]
}]
},
'active_player_id': 1
}
stanek = StateAsDict()
stanek.load_from_dict(dict)
fufu = SplendorGUI()
fufu.draw_state(stanek.to_state())
fufu.keep_window_open(300)
def dict_to_state(s_dict):
s_as_dict = StateAsDict()
s_as_dict.load_from_dict(s_dict)
state_to_return = s_as_dict.to_state()
state_to_return.active_player_id = 0
return state_to_return