def main(self, args):
"""
Train and save the DQN model, for the cartpole problem
:param args: (ArgumentParser) the input arguments
"""
#env = gym.make('CartPole-v1')
#model = DQN(MlpPolicy, env, verbose=1)
#model.load("cartpole_model.pkl")
model = DQN(env=env,
policy=CustomPolicy,
learning_rate=1e-3,
buffer_size=50000,
exploration_fraction=0.01,
exploration_final_eps=0.02,
verbose=1)
model.learn(total_timesteps=args.max_timesteps, callback=self.callback)
print("Saving model to cartpole_model.pkl")
model.save("cartpole_model.pkl")
#if __name__ == '__main__':
#parser = argparse.ArgumentParser(description="Train DQN on cartpole")
#parser.add_argument('--max-timesteps', default=100000000, type=int, help="Maximum number of timesteps")
#args = parser.parse_args()
#main(args)
def main(args):
"""
Train and save the DQN model, for the cartpole problem
:param args: (ArgumentParser) the input arguments
"""
# env = gym.make("CartPole-v0")
# model = DQN(
# env=env,
# policy=MlpPolicy,
# verbose=1,
# learning_rate=1e-3,
# buffer_size=50000,
# exploration_fraction=0.1,
# exploration_final_eps=0.02,
# tensorboard_log='./log',
# )
# model.learn(total_timesteps=args.max_timesteps, callback=callback)
# print("Saving model to cartpole_model.pkl")
# model.save("cartpole_model.pkl")
# env = Vrep_Env()
env = gym.make('vrep-v0')
model = DQN(
env=env,
gamma=0.95,
policy=MlpPolicy,
#policy=CustomPolicy,
verbose=1,
learning_rate=1e-4,
buffer_size=50000, #5000
train_freq=1,
learning_starts=100,
batch_size=64, # 32
checkpoint_freq=3000,
checkpoint_path='./model/',
target_network_update_freq=300,
prioritized_replay=True,
exploration_fraction=0.1,
exploration_final_eps=0.02,
tensorboard_log='./log',
)
# path = './model/'
# model = DQN.load(path+'bk2_16/cartpole_model6000.pkl', env, tensorboard_log='./log')
model.learn(total_timesteps=args.max_timesteps,
callback=callback,
log_interval=30)
print("Saving model to slab_installing_model.pkl")
model.save("slab_installing_model.pkl")
def train(env, fname):
env.setRender(False)
env.reset()
start = time.time()
model = DQN(
env=env,
policy=CustomPolicy,
learning_rate=1e-3,
buffer_size=50000,
exploration_fraction=0.1,
exploration_final_eps=0.02
)
model.learn(total_timesteps=STEPS, callback=callback)
# save trained model
model.save(fname)
print("Duration: %.1f" % ((time.time() - start)/60))
def main(args):
"""
Train and save the DQN model, for the cartpole problem
:param args: (ArgumentParser) the input arguments
"""
env = gym.make("CartPole-v0")
model = DQN(
env=env,
policy=MlpPolicy,
learning_rate=1e-3,
buffer_size=50000,
exploration_fraction=0.1,
exploration_final_eps=0.02,
)
model.learn(total_timesteps=args.max_timesteps, callback=callback)
print("Saving model to cartpole_model.pkl")
model.save("cartpole_model.pkl")
def main(args):
"""
Train and save the DQN model, for the mountain car problem
:param args: (ArgumentParser) the input arguments
"""
env = gym.make("MountainCar-v0")
# using layer norm policy here is important for parameter space noise!
model = DQN(policy=CustomPolicy,
env=env,
learning_rate=1e-3,
buffer_size=50000,
exploration_fraction=0.1,
exploration_final_eps=0.1,
param_noise=True)
model.learn(total_timesteps=args.max_timesteps)
print("Saving model to mountaincar_model.pkl")
model.save("mountaincar_model.pkl")