init evalCallback for training with sb3

examples/shields/rl/13_minigridsb.py

@@ -11,6 +11,7 @@ import time
 
 from utils import MiniGridShieldHandler, create_log_dir, ShieldingConfig, MiniWrapper
 from sb3utils import MiniGridSbShieldingWrapper, parse_sb3_arguments, ImageRecorderCallback, InfoCallback
+from stable_baselines3.common.callbacks import EvalCallback
 
 import os
 
@@ -33,27 +34,29 @@ def main():
     env = MiniWrapper(env)
     env = MiniGridSbShieldingWrapper(env, shield_handler=shield_handler, create_shield_at_reset=False, mask_actions=args.shielding == ShieldingConfig.Full)
     env = ActionMasker(env, mask_fn)
-    model = MaskablePPO("CnnPolicy", env, verbose=1, tensorboard_log=create_log_dir(args))
+    logDir = create_log_dir(args)
+    model = MaskablePPO("CnnPolicy", env, verbose=1, tensorboard_log=logDir)
 
+    evalCallback = EvalCallback(env, best_model_save_path=logDir,
+                                log_path=logDir, eval_freq=max(500,  int(args.steps/30)),
+                                deterministic=True, render=False)
     steps = args.steps
 
-    model.learn(steps,callback=[ImageRecorderCallback(), InfoCallback()], log_interval=1)
-
-
-    print("Learning done, hit enter")
-    input("")
-    vec_env = model.get_env()
-    obs = vec_env.reset()
-    terminated = truncated = False
-    while not terminated and not truncated:
-        action_masks = None
-        action, _states = model.predict(obs, action_masks=action_masks)
-        print(action)
-        obs, reward, terminated, truncated, info = env.step(action)
-        # action, _states = model.predict(obs, deterministic=True)
-        # obs, rewards, dones, info = vec_env.step(action)
-        vec_env.render("human")
-        time.sleep(0.2)
+    model.learn(steps,callback=[ImageRecorderCallback(), InfoCallback()])
+
+
+    #vec_env = model.get_env()
+    #obs = vec_env.reset()
+    #terminated = truncated = False
+    #while not terminated and not truncated:
+    #    action_masks = None
+    #    action, _states = model.predict(obs, action_masks=action_masks)
+    #    print(action)
+    #    obs, reward, terminated, truncated, info = env.step(action)
+    #    # action, _states = model.predict(obs, deterministic=True)
+    #    # obs, rewards, dones, info = vec_env.step(action)
+    #    vec_env.render("human")
+    #    time.sleep(0.2)