added saving for info - saves as tensors...need to consider waht is a…

…ctually required

cares_reinforcement_learning/util/Record.py

@@ -24,6 +24,7 @@ def __init__(self, glob_log_dir=None, log_dir=None, network=None, config=None) -
 
         self.train_data = pd.DataFrame()
         self.eval_data = pd.DataFrame()
+        self.info_data = pd.DataFrame()
 
         self.network = network
 
@@ -35,8 +36,9 @@ def __init__(self, glob_log_dir=None, log_dir=None, network=None, config=None) -
             with open(f'{self.directory}/config.yml', 'w') as outfile:
                 yaml.dump(config, outfile, default_flow_style=False)
 
-    def log_info(self, display=False, **logs):
-        pass # TODO implement logger for info from training the network e.g. loss rates etc etc
+    def log_info(self, info, display=False):
+        self.info_data = pd.concat([self.info_data, pd.DataFrame([info])], ignore_index=True)
+        self.save_data(self.info_data, "info", info, display=display)
 
     def log_train(self, display=False, **logs):
         self.log_count += 1

cares_reinforcement_learning/util/plotter.py

@@ -13,12 +13,12 @@
 def plot_data(plot_frame, title, label, x_label, y_label, directory, filename, display=True, close_figure=True):
     window_size = plot_frame["window_size"]
 
+    # TODO make font size a parameter
     plt.xlabel(x_label, fontsize=10)
     plt.ylabel(y_label, fontsize=10)
     plt.title(title, fontsize=10) 
 
     ax = sns.lineplot(data=plot_frame, x=plot_frame["steps"], y="avg", label=label)
-    # ax.set(xlabel=x_label, ylabel=y_label)
 
     Z = 1.960 # 95% confidence interval
     confidence_interval = Z * plot_frame["std_dev"] / np.sqrt(window_size)
@@ -122,8 +122,8 @@ def main():
         train_plot_frames.append(train_plot_frame)
         eval_plot_frames.append(eval_plot_frame)
 
-    plot_comparisons(train_plot_frames, f"{title}", labels, "Steps", "Average Reward", directory, "compare-train", True)
-    plot_comparisons(eval_plot_frames, f"{title}", labels, "Steps", "Average Reward", directory, "compare-eval", True)
+    plot_comparisons(train_plot_frames, f"{title}", labels, "Steps", "Average Reward", directory, f"{title}-compare-train", True)
+    plot_comparisons(eval_plot_frames, f"{title}", labels, "Steps", "Average Reward", directory, f"{title}-compare-eval", True)
 
 if __name__ == '__main__':
     main()

example/example_training_loops.py

@@ -105,10 +105,10 @@ def main():
 
     logging.info(f"Memory: {args['memory']}")
 
-    # Train the policy or value based approach
+    #create the record class - standardised results tracking
     record = Record(network=agent, config={'args': args})
+    # Train the policy or value based approach
     if args["algorithm"] == "PPO":
-        #create the record class
         ppe.ppo_train(env, agent, record, args)
         env = gym.make(env.spec.id, render_mode="human")
         ppe.evaluate_ppo_network(env, agent, args)

example/policy_example.py

@@ -95,8 +95,8 @@ def policy_based_train(env, agent, memory, record, args):
                     experience['next_state'],
                     experience['done']
                 ))
-                memory.update_priorities(experience['indices'], info)                
-                # TODO add saving info information from train_policy as seperate recording
+                memory.update_priorities(experience['indices'], info)
+                # record.log_info(info, display=False)
 
         if (total_step_counter+1) % number_steps_per_evaluation == 0:
             evaluate = True

example/ppo_example.py

@@ -91,7 +91,7 @@ def ppo_train(env, agent, record, args):
                 experience['done'],
                 experience['log_prob']
             ))
-            # TODO add saving info information from train_policy as seperate recording
+            # record.log_info(info, display=False)
 
         if (total_step_counter+1) % number_steps_per_evaluation == 0:
             evaluate = True

example/value_example.py

@@ -100,7 +100,7 @@ def value_based_train(env, agent, memory, record, args):
                     experience['done']
                 ))
                 memory.update_priorities(experience['indices'], info)
-                # TODO add saving info information from train_policy as seperate recording
+                # record.log_info(info, display=False)
 
         if (total_step_counter+1) % number_steps_per_evaluation == 0:
             evaluate = True