class HeuristicValuePolicy(EvaluationPolicy):
def __init__(self):
super().__init__(name='HeuristicValuePolicy')
self.value_function = ValueFunction()
def evaluate_state(self,
state: State,
list_of_actions: List[Action] = None) -> float:
inversed_state = StateAsDict(state).to_state()
inversed_state.change_active_player()
return self.value_function.evaluate(
state) - self.value_function.evaluate(inversed_state)
def evaluate_states(files_dir, dump_dir):
evaluator = ValueFunction()
list_of_files = os.listdir(files_dir)
for file_name in list_of_files:
with open(os.path.join(files_dir, file_name), 'rb') as f:
X, _ = pickle.load(f)
Y = []
for x in X:
state_to_eval = StateAsDict(x).to_state()
Y.append(evaluator.evaluate(state_to_eval))
del state_to_eval
with open(os.path.join(dump_dir, file_name), 'wb') as f:
pickle.dump((X, Y), f)
print(len(X))
del X
del Y
class ValueFunctionAgent(Agent):
def __init__(self):
super().__init__()
self.name = 'Value function Agent'
self.evaluator = ValueFunction()
def set_weights(self, weights):
self.evaluator.set_weights(weights)
def show_weights(self):
return self.evaluator.weights
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
class HeuraEvaluator(EvaluationPolicy):
def __init__(self):
super().__init__(name='Heura Value evaluator')
self.evaluator = ValueFunction()
def evaluate_state(self,
state: State,
list_of_actions: List[Action] = None) -> float:
#check if the state is terminal
if state.active_players_hand().number_of_my_points() >= POINTS_TO_WIN:
return -1
elif state.other_players_hand().number_of_my_points() >= POINTS_TO_WIN:
return 1
else:
return self.evaluator.evaluate(state)
def __init__(self):
super().__init__(name='HeuristicValuePolicy')
self.value_function = ValueFunction()
import pickle
import matplotlib.pyplot as plt
from nn_models.value_function_heura.value_function import ValueFunction
with open('/home/tomasz/ML_Research/splendor/gym-splendor/supervised_data/test0/valid_epochs/epoch_0.pickle', 'rb') as f:
X, _ = pickle.load(f)
vf = ValueFunction()
Y = [vf.evaluate(st) for st in X]
print(len(Y))
print(min(Y))
print(max(Y))
plt.hist(Y, bins=60)
plt.show()
def __init__(self):
super().__init__()
self.name = 'Value function Agent'
self.evaluator = ValueFunction()
def __init__(self):
super().__init__(name='Heura Value evaluator')
self.evaluator = ValueFunction()