[ci skip] added noise to the DDPG example in HER

Stable-Baselines-Team · Sep 17, 2018 · 34e00d2 · 34e00d2
1 parent c77cb85
commit 34e00d2
Showing 1 changed file with 4 additions and 1 deletion.
diff --git a/docs/modules/her.rst b/docs/modules/her.rst
@@ -28,11 +28,14 @@ Train a HER agent on `MountainCarContinuous-v0`.
   from stable_baselines.common.vec_env import DummyVecEnv
   from stable_baselines.her.reward_class import ProximalReward
   from stable_baselines.her.utils import stack_obs_goal
+  from stable_baselines.ddpg.noise import AdaptiveParamNoiseSpec
   from stable_baselines import DDPG, HER
 
   env = DummyVecEnv([lambda: gym.make('MountainCarContinuous-v0')])  # The algorithms require a vectorized environment to run
 
-  model = HER(DDPG, 'MlpPolicy', env, ProximalReward(eps=0.1))  # define the reward function for HER
+  param_noise = AdaptiveParamNoiseSpec(initial_stddev=float(0.2), desired_action_stddev=float(0.2))
+
+  model = HER(DDPG, 'MlpPolicy', env, ProximalReward(eps=0.1), param_noise=param_noise)  # define the reward function for HER
   model.learn(total_timesteps=25000)
   model.save("her_dqn_mountaincar")