fix bug

PPO_RND/pytorch/ppo_rnd_frozen_notslippery_pytorch.py

@@ -484,7 +484,7 @@ def run_inits_episode(env, agent, state_dim, render, n_init_episode):
  env.reset()
 
  agent.updateObsNormalizationParam(agent.obs_memory.observations)
- agent.memory.clearObs()
+ agent.obs_memory.clearMemory()
 
  return agent
 
@@ -547,6 +547,7 @@ def main():
  n_eps_update = 5 # How many episode before you update the PPO. Recommended set to 5 for Discrete
  n_plot_batch = 100000000 # How many episode you want to plot the result
  n_episode = 100000 # How many episode you want to run
+ n_init_episode = 256
  n_saved = 10 # How many episode to run before saving the weights
 
  policy_kl_range = 0.0008 # Recommended set to 0.0008 for Discrete
@@ -587,6 +588,13 @@ def main():
 
  t_updates = 0
 
+ #############################################
+
+ if training_mode:
+ agent = run_inits_episode(env, agent, state_dim, render, n_init_episode)
+
+ #############################################
+
  for i_episode in range(1, n_episode + 1):
  total_reward, time, t_updates = run_episode(env, agent, state_dim, render, training_mode, t_updates, n_step_update)
  print('Episode {} \t t_reward: {} \t time: {} \t '.format(i_episode, total_reward, time))

PPO_RND/pytorch/ppo_rnd_pytorch.py

@@ -483,7 +483,7 @@ def run_inits_episode(env, agent, state_dim, render, n_init_episode):
  env.reset()
 
  agent.updateObsNormalizationParam(agent.obs_memory.observations)
- agent.memory.clearObs()
+ agent.obs_memory.clearMemory()
 
  return agent
 
@@ -533,6 +533,7 @@ def main():
  n_eps_update = 5 # How many episode before you update the PPO. Recommended set to 5 for Discrete
  n_plot_batch = 100000000 # How many episode you want to plot the result
  n_episode = 100000 # How many episode you want to run
+ n_init_episode = 1024
  n_saved = 10 # How many episode to run before saving the weights
 
  policy_kl_range = 0.0008 # Recommended set to 0.0008 for Discrete
@@ -573,6 +574,13 @@ def main():
 
  t_updates = 0
 
+ #############################################
+
+ if training_mode:
+ agent = run_inits_episode(env, agent, state_dim, render, n_init_episode)
+
+ #############################################
+
  for i_episode in range(1, n_episode + 1):
  total_reward, time, t_updates = run_episode(env, agent, state_dim, render, training_mode, t_updates, n_step_update)
  print('Episode {} \t t_reward: {} \t time: {} \t '.format(i_episode, total_reward, time))

PPO_RND/tensorflow 2/ppo_frozenlake_notslippery_tensorflow.py

@@ -451,7 +451,7 @@ def run_inits_episode(env, agent, state_dim, render, n_init_episode):
  env.reset()
 
  agent.updateObsNormalizationParam(agent.obs_memory.observations)
- agent.memory.clearObs()
+ agent.obs_memory.clearMemory()
 
  return agent
 
@@ -514,6 +514,7 @@ def main():
  n_eps_update = 5 # How many episode before you update the PPO. Recommended set to 5 for Discrete
  n_plot_batch = 100000000 # How many episode you want to plot the result
  n_episode = 100000 # How many episode you want to run
+ n_init_episode = 256
  n_saved = 10 # How many episode to run before saving the weights
 
  policy_kl_range = 0.0008 # Recommended set to 0.0008 for Discrete
@@ -553,6 +554,13 @@ def main():
  batch_times = []
 
  t_updates = 0
+
+ #############################################
+
+ if training_mode:
+ agent = run_inits_episode(env, agent, state_dim, render, n_init_episode)
+
+ #############################################
 
  for i_episode in range(1, n_episode + 1):
  total_reward, time, t_updates = run_episode(env, agent, state_dim, render, training_mode, t_updates, n_step_update)

PPO_RND/tensorflow 2/ppo_rnd_tensorflow.py

@@ -451,7 +451,7 @@ def run_inits_episode(env, agent, state_dim, render, n_init_episode):
  env.reset()
 
  agent.updateObsNormalizationParam(agent.obs_memory.observations)
- agent.memory.clearObs()
+ agent.obs_memory.clearMemory()
 
  return agent
 
@@ -514,6 +514,7 @@ def main():
  n_eps_update = 5 # How many episode before you update the PPO. Recommended set to 5 for Discrete
  n_plot_batch = 100000000 # How many episode you want to plot the result
  n_episode = 100000 # How many episode you want to run
+ n_init_episode = 1024
  n_saved = 10 # How many episode to run before saving the weights
 
  policy_kl_range = 0.0008 # Recommended set to 0.0008 for Discrete
@@ -554,6 +555,13 @@ def main():
 
  t_updates = 0
 
+ #############################################
+
+ if training_mode:
+ agent = run_inits_episode(env, agent, state_dim, render, n_init_episode)
+
+ #############################################
+
  for i_episode in range(1, n_episode + 1):
  total_reward, time, t_updates = run_episode(env, agent, state_dim, render, training_mode, t_updates, n_step_update)
  print('Episode {} \t t_reward: {} \t time: {} \t '.format(i_episode, total_reward, time))