Allow generalization training with specified arguments of min_reward and min_lesson_length

ml-agents/mlagents/trainers/learn.py

             docker_target_name=docker_target_name
         )
 
+    if sampler is not None:
+        sampler = load_config(sampler)
+
-    reset_param_dict = load_config(reset_param_dict_path)
+    # reset_param_dict = load_config(reset_param_dict_path)
     env_factory = create_environment_factory(
         env_path,
         docker_target_name,

ml-agents/mlagents/trainers/trainer_controller.py

             lessons_incremented = self.meta_curriculum.increment_lessons(
                 self._get_measure_vals(), reward_buff_sizes=reward_buff_sizes
             )
+        elif (self.sampler_manager is not None):
+            reward_buff_mean = {
+                k: np.mean(t.reward_buffer) if len(t.reward_buffer) > 0 else 0.0 for (k, t) in self.trainers.items()
+            }
+            lessons_incremented = {
+                k: (t > self.min_reward) for (k, t) in reward_buff_mean.items()
+            }
         else:
             lessons_incremented = {}