import time
# from stable_baselines.common.policies import MlpPolicy
from stable_baselines.common.vec_env import DummyVecEnv
from stable_baselines import DDPG
from stable_baselines.ddpg.policies import MlpPolicy
if __name__ == '__main__':
# Create and wrap the environment
# env = gym.make('game-stock-exchange-v0')
env = gym.make('game-stock-exchange-continuous-v0')
env = DummyVecEnv([lambda: env])
model = DDPG(MlpPolicy, env, verbose=1)
# model = A2C.load("a2c_gym_exchange_continuous", env=env)
model.learning_rate = 1e-7
# Train the agent
model.learn(total_timesteps=100000) # default=1000000
# Save the agent
model.save("ddpg_gym_exchange_continuous")
# del model # delete trained model to demonstrate loading
# Load the trained agent
model = DDPG.load("ddpg_gym_exchange_continuous", env=env)
# Enjoy trained agent
obs = env.reset()
actions = Counter()
pnl = defaultdict(float)
str([
my_signal_rate, my_signal_repetitions, my_step_limit, lr_start,
lr_end, timesteps
]))
f.close()
except:
print("envparameters couldn't be saved. They are:" + str([
my_signal_rate, my_signal_repetitions, my_step_limit, lr_start, lr_end,
timesteps
]))
lr_update_interval = 500
model_save_interval = 20000
modulo_number = model_save_interval / lr_update_interval
i = 0
while (i <= (timesteps / lr_update_interval)):
# linear: model.learning_rate = lr_start-(lr_stepsize*(i+pretraining_iterations))
# log:
model.learning_rate = lr_start * 0.5**((i * lr_update_interval) *
(10 / timesteps))
# static: model.learning_rate = static_learning_rate
model.learn(total_timesteps=lr_update_interval,
tb_log_name=name,
log_interval=10,
reset_num_timesteps=False)
if (i % modulo_number == (modulo_number - 1)):
model.save("/media/ryuga/TOSHIBA EXT/BA/Models/" + name + "_" +
str(i / modulo_number))
i += 1