def __init__(self,
policy_func: types.MethodType = None,
trained_path: str = "./_trained/Q_Learning/"):
self.policy_func = policy_func if policy_func else policy.ActionPolicies(
).greedy_maximum
self.trained_path = trained_path
if not os.path.exists(trained_path):
raise RuntimeError("[Error] Trained Path NOT Found")
# Model Related Initialization
self.q_table = None
self.epsilon = None
def __init__(self,
train_epoch: int = 10000,
policy_func: types.MethodType = None,
update_func: types.MethodType = None,
output_path: str = "./_trained/Q_Learning/",
progress_bar: bool = True):
self.train_epoch = train_epoch
self.env_obj = environment.Easy21Env()
self.policy_func = policy_func if policy_func else policy.ActionPolicies(
).greedy_epsilon
self.update_func = update_func if update_func \
else update.UpdateQTable().q_function # learning_rate=0.1, discount_factor=0.5
self.progress_bar = progress_bar
_ = check_output_path(output_path=output_path)
self.output_path = output_path
evaluate_result_path, "evaluation_report_%s" %
datetime.now().strftime("%Y%m%d%H%M%S")), "w") as f:
for _model in models_result:
f.write("Filename:%s\t"
"Epoch:%d\tLearning Rate:%f\t"
"Discount Factor:%f\tEpsilon:%f\t"
"WIN:%d\tTIE:%d\tLOSE:%d\tERR:%d\tALL:%d\n" %
(_model["filename"], _model["setting"]["epoch"],
_model["setting"]["learning rate"],
_model["setting"]["discount factor"],
_model["setting"]["epsilon"], _model["result"][1],
_model["result"][0], _model["result"][-1],
_model["result"]["err"], evaluate_rounds))
if _model["result"][1] > max_win_cnt:
max_win_cnt = _model["result"][1]
max_win_rate_setting = _model["setting"]
print("\nEvaluation Done\n")
print("[Max Win Count]", max_win_cnt)
print("[Max Count Setting]", max_win_rate_setting)
if "__main__" == __name__:
eval_env_obj = evaluate.EvaluateEnv(show_details=False)
policy_func = policy.ActionPolicies().greedy_epsilon
ql_env_obj = evaluate.QLearningEnv(**{"policy_func": policy_func})
eval_param(trained_model_path="../../_trained/", evaluate_rounds=1000)