for i in range(len(state_list)):
dealerFirstCard = state_list[i].dealerFirstCard().num*state_list[i].dealerFirstCard().sign
sumOfmine = state_list[i].sumOfmine()
action = action_list[i]
# s = self.Q[dealerFirstCard,sumOfmine,action]*self.N[dealerFirstCard,sumOfmine,action] + reward
self.N[dealerFirstCard,sumOfmine,action] += 1
alpha = 1/self.N[dealerFirstCard,sumOfmine,action]
self.Q[dealerFirstCard,sumOfmine,action] = self.Q[dealerFirstCard,sumOfmine,action] + alpha*(reward - self.Q[dealerFirstCard,sumOfmine,action])
# self.Q[dealerFirstCard,sumOfmine,action] = s/self.N[dealerFirstCard,sumOfmine,action]
return
if __name__ == '__main__':
env = env()
agent = Egreedy_MC_agent()
reward = 0
iteration = 50000
reward_list = []
for i in range(iteration):
reward = 0
state_list = []
action_list = []
if i == 0:
state = env.init_state()
else:
state = env.Restart()
epsilon=1, # 随机选择的初始概率
epsilon_decay=1E-5, # 随机选择的概率decay
batch_size=256,
lr=0.01,
T=3,
mem_size=50000)
scores = []
for i in range(num_eposides):
# 设定Env
# graph_size = np.random.randint(50, 100)
# seed_size = np.random.randint(10, 30)
graph_size = 100
seed_size = 25
Env = env(graph_size=graph_size, seed_size=seed_size, edge_weight=0.1,
random_edge_weight=False, network_model="BA")
edge_index, edge_weight, x, mu, done = Env.reset()
graph = Data(edge_index = torch.LongTensor(edge_index),
edge_weight = torch.Tensor(edge_weight),
mu = torch.Tensor(mu),
x = torch.Tensor(x))
# to be stored
graph_former_steps = []
graph_later_steps = []
action_steps = []
reward_steps = []
done_steps = []
steps_cntr = 0
# running eposide
while not done:
def MSS(edge_index):
"""Networkx Greedy for MSS"""
out_edge = edge_index.numpy()
g = create_graph(out_edge)
new_g = nx.Graph()
for (u,v) in g.edges():
new_g.add_edge(u,v)
new_g = complement(new_g)
SS_list = list(nx.find_cliques(new_g))
MSS_list = max(SS_list,key=len)
return MSS_list
if __name__ == "__main__":
num_episodes = 20000 #! Number could be changed
env = env(train_size=40,i=1,episodes=num_episodes) #!
n_step = 4
agent = Agent()
scores = []
loss_list = []
"""
Please note that after adding this Greedy Algorithm, your running time will be much longer!!!
"""
Greedy_filename = "greedy.txt"
file_greedy = open(Greedy_filename,"w")
score_filename = "score.txt"#!
file_score = open(score_filename,"w")#!
for i in tqdm(range(num_episodes)):
score = 0
graph = env.reset(i,num_episodes)#!
file_greedy.write("Episode: {:<4}, size_greedy: {:<4}, node_num:{:<4}, stable_set:{}\n".