if __name__ == "__main__":
max_round = 30
file_path = '000065.SZ_NormalData.csv'
df = pd.read_csv(file_path)
df = df.sort_values('trade_date', ascending=True)
df = df.iloc[22:].reset_index(drop=True) # 去除前几天没有均线信息
env = stock(df.iloc[0:1500])
RL = DeepQNetwork(
env.n_actions,
env.n_features,
learning_rate=0.02,
reward_decay=0.9,
e_greedy=0.9,
replace_target_iter=200,
memory_size=4000,
batch_size=512,
# output_graph=True
)
run(max_round)
# env = stock(df)
# env = BackTest(env, show_log=True)
# env.draw('trade.png', 'profit.png')
env = stock(df.iloc[1500:].reset_index(drop=True))
env = BackTest(env, show_log=True)
env.draw('trade1.png', 'profit1.png')
env = BackTest(env, show_log=True, my_trick=True)
import gym
from RL_brain2 import DeepQNetwork
env = gym.make('SpaceInvaders-v0')
env = env.unwrapped
print("action_space:", env.action_space)
print("observation_space:", env.observation_space)
print("observation_space.high:", env.observation_space.high)
print("observation_space.low:", env.observation_space.low)
RL = DeepQNetwork(n_actions=env.action_space.n,
n_features=env.observation_space.shape[0] * env.observation_space.shape[1] * env.observation_space.shape[2],
learning_rate=0.01, e_greedy=0.9,
replace_target_iter=100, memory_size=2000,
e_greedy_increment=0.001,)
total_steps = 0
ep_rhistory = []
for i_episode in range(500):
observation = env.reset()
ep_r = 0
while True:
# env.render()
import tkinter as tk
from env import crossing
from visual import Visual
np.set_printoptions(threshold=np.inf)
#print(env.observation_space.shape[0])
parser = argparse.ArgumentParser(description='Train or test neural net motor controller.')
parser.add_argument('--train', dest='train', action='store_true', default=False)
parser.add_argument('--test', dest='test', action='store_true', default=True)
args = parser.parse_args()
RL = DeepQNetwork(n_actions=4, #2*2
#n_features=env.observation_space.shape[0],
n_features=10, #2*5
learning_rate=0.01, e_greedy=0.9,
replace_target_iter=100, memory_size=2000,
e_greedy_increment=0.001,)
def road_map():
cross1=crossing(light_state=0,q_states=[0,0,0,1])
cross2=crossing(light_state=0,q_states=[0,0,1,0])
step_set=[]
reward_set=[]
#print(env.observation_space.shape[0])
parser = argparse.ArgumentParser(
description='Train or test neural net motor controller.')
parser.add_argument('--train', dest='train', action='store_true', default=True)
# parser.add_argument('--test', dest='test', action='store_true', default=True)
args = parser.parse_args()
# size of crossroads grid
grid_x = 4
grid_y = 4
RL = DeepQNetwork(
n_actions=2**(grid_x * grid_y), #0,1 for each crossroad
n_features=5 *
(grid_x * grid_y), #2*5 (5 = 4 numbers of cars + 1 light state)
learning_rate=0.01,
e_greedy=0.9,
replace_target_iter=100,
memory_size=2000,
e_greedy_increment=0.001,
)
x = []
y = []
for i in range(grid_x):
x.append(i + 1)
for i in range(grid_y):
y.append(i + 1)
#property of visualization
times = 100 #interval: crossroad & crossroad
bias = 6 #distance: light & center of crossroad
print 'success =', success
break
# swap observation
observation = observation_
step += 1
# end of game
print('game over')
plt.plot(np.arange(episode_number), rr_episode, '.')
plt.ylabel('reward')
plt.xlabel('training episode')
plt.show()
plt.plot(np.arange(episode_number), step_episode, '.')
plt.ylabel('step')
plt.xlabel('training episode')
plt.show()
if __name__ == "__main__":
# maze game
env = UR5()
RL = DeepQNetwork(4,
12,
learning_rate=0.00001,
reward_decay=0.5,
e_greedy=1,
replace_target_iter=3,
memory_size=4000,
output_graph=False)
run()
RL.plot_cost()