pre major rl eval change

src/active_bo_msgs/msg/ActiveBOResponse.msg

@@ -1,4 +1,4 @@
-float32[] best_policy
-float32[] best_weights
-float32[] reward_mean
-float32[] reward_std
+float64[] best_policy
+float64[] best_weights
+float64[] reward_mean
+float64[] reward_std

src/active_bo_msgs/msg/ActiveBOState.msg

@@ -1,4 +1,4 @@
 uint16 current_run
 uint16 current_episode
-float32 best_reward
-float32 last_user_reward
+float64 best_reward
+float64 last_user_reward

src/active_bo_msgs/msg/ActiveRL.msg

@@ -1,5 +1,5 @@
 string env
 uint32 seed
 bool final_run
-float32[] policy
-float32[] weights
+float64[] policy
+float64[] weights

src/active_bo_msgs/msg/ActiveRLResponse.msg

@@ -1,3 +1,3 @@
-float32[] weights
+float64[] weights
 uint16 final_step
-float32 reward
+float64 reward

src/active_bo_ros/active_bo_ros/BayesianOptimization/BayesianOptimization.py

@@ -51,7 +51,7 @@ class BayesianOptimization:
         for i in range(len(policy)):
             action = policy[i]
             action_clipped = action.clip(min=-1.0, max=1.0)
-            output = self.env.step(action_clipped.astype(np.float32))
+            output = self.env.step(action_clipped.astype(np.float64))
 
             env_reward += output[1]
             done = output[2]
@@ -71,8 +71,8 @@ class BayesianOptimization:
         self.env.reset(seed=seed)
         self.reset_bo()
 
-        self.X = np.zeros((self.nr_init, self.nr_policy_weights), dtype=np.float32)
-        self.Y = np.zeros((self.nr_init, 1), dtype=np.float32)
+        self.X = np.zeros((self.nr_init, self.nr_policy_weights), dtype=np.float64)
+        self.Y = np.zeros((self.nr_init, 1), dtype=np.float64)
 
         for i in range(self.nr_init):
             self.policy_model.random_policy()
@@ -122,34 +122,34 @@ class BayesianOptimization:
         return x_next
 
     def eval_new_observation(self, x_next, seed=None):
-        self.policy_model.weights = x_next
+        self.policy_model.weights = np.around(x_next, decimals=8)
         policy = self.policy_model.rollout()
 
         reward, step_count = self.runner(policy, seed=seed)
 
-        self.X = np.vstack((self.X, x_next), dtype=np.float32)
-        self.Y = np.vstack((self.Y, reward), dtype=np.float32)
+        self.X = np.vstack((self.X, np.around(x_next, decimals=8)), dtype=np.float64)
+        self.Y = np.vstack((self.Y, reward), dtype=np.float64)
 
         self.GP.fit(self.X, self.Y)
 
         if self.episode == 0:
             self.best_reward[0] = np.max(self.Y)
         else:
-            self.best_reward = np.vstack((self.best_reward, np.max(self.Y)), dtype=np.float32)
+            self.best_reward = np.vstack((self.best_reward, np.max(self.Y)), dtype=np.float64)
 
         self.episode += 1
         return step_count
 
     def add_new_observation(self, reward, x_new):
-        self.X = np.vstack((self.X, x_new), dtype=np.float32)
-        self.Y = np.vstack((self.Y, reward), dtype=np.float32)
+        self.X = np.vstack((self.X, np.around(x_new, decimals=8)), dtype=np.float64)
+        self.Y = np.vstack((self.Y, reward), dtype=np.float64)
 
         self.GP.fit(self.X, self.Y)
 
         if self.episode == 0:
             self.best_reward[0] = np.max(self.Y)
         else:
-            self.best_reward = np.vstack((self.best_reward, np.max(self.Y)), dtype=np.float32)
+            self.best_reward = np.vstack((self.best_reward, np.max(self.Y)), dtype=np.float64)
 
         self.episode += 1
 

src/active_bo_ros/active_bo_ros/PolicyModel/GaussianRBFModel.py

@@ -24,7 +24,7 @@ class GaussianRBF:
         self.policy = np.zeros((self.nr_steps, 1))
 
     def random_policy(self):
-        self.weights = self.rng.uniform(self.low, self.upper, self.nr_weights)
+        self.weights = np.around(self.rng.uniform(self.low, self.upper, self.nr_weights), decimals=8)
 
     def rollout(self):
         self.policy = np.zeros((self.nr_steps, 1))

src/active_bo_ros/active_bo_ros/ReinforcementLearning/CartPole.py

@@ -101,15 +101,15 @@ class CartPoleEnv(gym.Env[np.ndarray, Union[int, np.ndarray]]):
         high = np.array(
             [
                 self.x_threshold * 2,
-                np.finfo(np.float32).max,
+                np.finfo(np.float64).max,
                 self.theta_threshold_radians * 2,
-                np.finfo(np.float32).max,
+                np.finfo(np.float64).max,
             ],
-            dtype=np.float32,
+            dtype=np.float64,
         )
 
-        self.action_space = Box(low=-1.0, high=1.0, shape=(1,), dtype=np.float32)
-        self.observation_space = spaces.Box(-high, high, dtype=np.float32)
+        self.action_space = Box(low=-1.0, high=1.0, shape=(1,), dtype=np.float64)
+        self.observation_space = spaces.Box(-high, high, dtype=np.float64)
 
         self.render_mode = render_mode
 
@@ -185,7 +185,7 @@ class CartPoleEnv(gym.Env[np.ndarray, Union[int, np.ndarray]]):
         if self.render_mode == "human":
             self.render()
 
-        return np.array(self.state, dtype=np.float32), reward, terminated, False, {}
+        return np.array(self.state, dtype=np.float64), reward, terminated, False, {}
 
     def reset(
         self,
@@ -204,7 +204,7 @@ class CartPoleEnv(gym.Env[np.ndarray, Union[int, np.ndarray]]):
 
         if self.render_mode == "human":
             self.render()
-        return np.array(self.state, dtype=np.float32), {}
+        return np.array(self.state, dtype=np.float64), {}
 
     def render(self):
         if self.render_mode is None:

src/active_bo_ros/active_bo_ros/active_bo_topic.py

@@ -329,13 +329,13 @@ class ActiveBOTopic(Node):
             if self.BO is not None and self.BO.Y is not None:
                 self.best_reward = np.max(self.BO.Y)
 
-            state_msg = ActiveBOState()
-            state_msg.current_run = self.current_run + 1 if self.current_run < self.bo_runs else self.bo_runs
-            state_msg.current_episode = self.current_episode + 1 \
-                if self.current_episode < self.bo_episodes else self.bo_episodes
-            state_msg.best_reward = float(self.best_reward)
-            state_msg.last_user_reward = float(self.rl_reward)
-            self.state_pub.publish(state_msg)
+                state_msg = ActiveBOState()
+                state_msg.current_run = self.current_run + 1 if self.current_run < self.bo_runs else self.bo_runs
+                state_msg.current_episode = self.current_episode + 1 \
+                    if self.current_episode < self.bo_episodes else self.bo_episodes
+                state_msg.best_reward = float(self.best_reward)
+                state_msg.last_user_reward = float(self.rl_reward)
+                self.state_pub.publish(state_msg)
 
 
 def main(args=None):

src/active_bo_ros/active_bo_ros/active_rl_topic.py

@@ -86,7 +86,7 @@ class ActiveRLService(Node):
     def active_rl_callback(self, msg):
         self.rl_env = msg.env
         self.rl_seed = msg.seed
-        self.rl_policy = np.array(msg.policy, dtype=np.float32)
+        self.rl_policy = np.array(msg.policy, dtype=np.float64)
         self.rl_weights = msg.weights
         self.final_run = msg.final_run
 
@@ -112,7 +112,7 @@ class ActiveRLService(Node):
         self.eval_weights = None
 
     def active_rl_eval_callback(self, msg):
-        self.eval_policy = np.array(msg.policy, dtype=np.float32)
+        self.eval_policy = np.array(msg.policy, dtype=np.float64)
         self.eval_weights = msg.weights
 
         self.get_logger().info('Active RL Eval: Responded!')
@@ -122,7 +122,7 @@ class ActiveRLService(Node):
     def next_image(self, policy):
         action = policy[self.rl_step]
         action_clipped = action.clip(min=-1.0, max=1.0)
-        output = self.env.step(action_clipped.astype(np.float32))
+        output = self.env.step(action_clipped.astype(np.float64))
 
         self.rl_reward += output[1]
         done = output[2]

src/active_bo_ros/active_bo_ros/rl_service.py

@@ -27,7 +27,7 @@ class RLService(Node):
 
         reward = 0
         step_count = 0
-        policy = np.array(request.policy, dtype=np.float32)
+        policy = np.array(request.policy, dtype=np.float64)
         rl_env = request.env
 
         if rl_env == "Mountain Car":
@@ -46,7 +46,7 @@ class RLService(Node):
         for i in range(len(policy)):
             action = policy[i]
             action_clipped = action.clip(min=-1.0, max=1.0)
-            output = self.env.step(action_clipped.astype(np.float32))
+            output = self.env.step(action_clipped.astype(np.float64))
 
             reward += output[1]
             done = output[2]