remove defender weight replay causing issues with dtap

setter.py

@@ -256,10 +256,10 @@ def __init__(self, mode, end_object_choice=None, simulator=False):
             for i in range(4):
                 self.setters.append(
                     WeightedSampleSetter(
-                        (AugmentSetter(ReplaySetter(double_tap_replays[i], defender_front_goal_weight=0.2,
+                        (AugmentSetter(ReplaySetter(double_tap_replays[i], defender_front_goal_weight=0,
                                                    random_boost=True, print_choice=True),
                         True, False, False),
-                         AugmentSetter(ReplaySetter(easy_double_tap_replays[i], defender_front_goal_weight=0.2,
+                         AugmentSetter(ReplaySetter(easy_double_tap_replays[i], defender_front_goal_weight=0,
                                                     random_boost=True, print_choice=True),
                                        True, False, False),
                          ), (0.2, 0.8))

sim_tester.py

@@ -12,6 +12,7 @@
 setter = AugmentSetter(ReplaySetter("replays/double_tap_1v1.npy", defender_front_goal_weight=0,
                                     random_boost=True, print_choice=True, override=None, index=index),
                        True, False, False)
+# setter = ReplaySetter("replays/bad_1v1_doubletap_state.npy")
 parser = CoyoteAction()
 obs = TestObs()
 reward = TestReward()
@@ -36,21 +37,22 @@
     done = False
     steps = 0
     while not done:
-        actions = np.asarray((np.asarray([0]), np.asarray([np.random.randint(0, 373)])))
+        #actions = np.asarray((np.asarray([0]), np.asarray([np.random.randint(0, 373)])))
         # actions = np.asarray(np.asarray([0],))
+        actions = np.asarray((np.asarray([0]), np.asarray([0])))
         new_obs, reward, done, state = env.step(actions)
         obs = new_obs
         if np.isnan(obs).any():
             # self.actions.reverse()
             # self.states.reverse()
-            print(f"There is a nan in the obs. Printing states")
+            print(f"There is a nan in the obs. Printing states after {steps} steps")
             if index[0] not in bad_list:
                 bad_list.append(index[0])
             # input()
             break
         steps += 1
     total_steps += steps
     print(bad_list)
-    # print(f"completed {steps} steps. Starting new episode. Done {total_steps} total steps")
+    print(f"completed {steps} steps. Starting new episode. Done {total_steps} total steps")
 
 

sim_tester_minimal.py

@@ -0,0 +1,40 @@
+import rlgym_sim
+# from CoyoteParser import CoyoteAction
+from rlgym_tools.extra_state_setters.replay_setter import ReplaySetter
+# from rlgym_tools.extra_state_setters.augment_setter import AugmentSetter
+# from test_files.test_obs import TestObs
+# from test_files.test_reward import TestReward
+import numpy as np
+from rlgym_sim.utils.terminal_conditions.common_conditions import GoalScoredCondition, TimeoutCondition
+
+bad_list = []
+index = {}
+setter = ReplaySetter("replays/bad_1v1_doubletap_state.npy")
+# obs = TestObs()
+# reward = TestReward()
+terminals = [GoalScoredCondition(), TimeoutCondition(100)]
+
+env = rlgym_sim.make(tick_skip=4, spawn_opponents=True, state_setter=setter, copy_gamestate_every_step=True,
+                     terminal_conditions=terminals)
+
+total_steps = 0
+
+while True:
+    obs = env.reset()
+    done = False
+    steps = 0
+    while not done:
+        # actions = np.asarray((np.asarray([0]), np.asarray([np.random.randint(0, 373)])))
+        # actions = np.asarray(np.asarray([0],))
+        # actions = np.asarray([0] * 8), np.asarray([0] * 8)
+        actions = np.asarray([1, 0.5, 0.5, 0.5, 0, 0, 1, 0]), np.asarray([1, 0.5, 0.5, 0.5, 0, 0, 1, 0])
+        new_obs, reward, done, state = env.step(actions)
+        obs = new_obs
+        if np.isnan(obs).any():
+            print(f"There is a nan in the obs after {steps} steps.")
+            break
+        steps += 1
+    total_steps += steps
+    print(f"completed {steps} steps. Starting new episode. Done {total_steps} total steps")
+
+