updated config

Author: tarun018

common/experience.py

@@ -40,7 +40,7 @@ def add(self, obs_t, action, state_t, reward, obs_tp1, done):
         self._states[self._next_idx] = state_t
 
         self._next_idx = (self._next_idx + 1) % self._maxsize
-        print(self._curr_size)
+        # print(self._curr_size)
 
     def _get(self, idx):
         return self._obses[idx], self._actions[idx], self._states[idx], self._rewards[idx], self._next_obses[idx], self._dones[idx]

configs/whirl_baselines/iql_MMM2.yaml

@@ -51,7 +51,7 @@ params:
     gamma: 0.99
     learning_rate: 0.0005
     steps_per_epoch: 10
-    batch_size: 2
+    batch_size: 64
     epsilon: 1.0
     min_epsilon: 0.05
     epsilon_decay_frames: 100000
@@ -60,14 +60,14 @@ params:
     name: MMM2
     is_double: True
     score_to_win: 20
-    num_steps_fill_buffer: 10
+    num_steps_fill_buffer: 10000
     replay_buffer_type: 'normal'
     replay_buffer_size: 1000000
     priority_beta: 0.4
     priority_alpha: 0.6
     beta_decay_frames: 100000
     max_beta: 1
-    steps_num: 1
+    steps_num: 10
 #    episodes_to_log: 10
     atoms_num: 1
     games_to_track: 20

configs/whirl_baselines/vdn_3s5z_vs_3s6z.yaml

@@ -57,10 +57,10 @@ params:
     epsilon_decay_frames: 100000
     num_epochs_to_copy: 10000
     env_name:  smac_cnn
-    name: 3s5z_vs_3s6z
+    name: MMM2
     is_double: True
     score_to_win: 20
-    num_steps_fill_buffer: 100000
+    num_steps_fill_buffer: 10000
     replay_buffer_type: 'normal'
     replay_buffer_size: 1000000
     priority_beta: 0.4

configs/whirl_baselines/vdn_3s_vs_5z.yaml

@@ -57,10 +57,10 @@ params:
     epsilon_decay_frames: 100000
     num_epochs_to_copy: 10000
     env_name:  smac_cnn
-    name: 3s_vs_5z
+    name: MMM2
     is_double: True
     score_to_win: 20
-    num_steps_fill_buffer: 100000
+    num_steps_fill_buffer: 10000
     replay_buffer_type: 'normal'
     replay_buffer_size: 1000000
     priority_beta: 0.4

configs/whirl_baselines/vdn_MMM2.yaml

@@ -60,7 +60,7 @@ params:
     name: MMM2
     is_double: True
     score_to_win: 20
-    num_steps_fill_buffer: 1000
+    num_steps_fill_buffer: 10000
     replay_buffer_type: 'normal'
     replay_buffer_size: 1000000
     priority_beta: 0.4