backup

environments/factory/renderer.py

@@ -5,7 +5,6 @@ from pathlib import Path
 from collections import deque
 import pygame
 
-
 @dataclass
 class Entity:
     name: str

environments/factory/simple_factory.py

@@ -8,9 +8,8 @@ import numpy as np
 from environments.factory.base_factory import BaseFactory, AgentState
 from environments import helpers as h
 
-from environments.factory.renderer import Renderer
-from environments.factory.renderer import Entity
 from environments.logging.monitor import MonitorCallback
+from environments.factory.renderer import Renderer, Entity
 
 DIRT_INDEX = -1
 
@@ -33,7 +32,7 @@ class SimpleFactory(BaseFactory):
     def _is_clean_up_action(self, action):
         return self.action_space.n - 1 == action
 
-    def __init__(self, *args, dirt_properties: DirtProperties, verbose=False, **kwargs):
+    def __init__(self, *args, dirt_properties: DirtProperties, verbose=False, force_skip_render=False, **kwargs):
         self._dirt_properties = dirt_properties
         self.verbose = verbose
         self.max_dirt = 20
@@ -42,6 +41,7 @@ class SimpleFactory(BaseFactory):
         self.renderer = None  # expensive - dont use it when not required !
 
     def render(self):
+
         if not self.renderer:  # lazy init
             height, width = self.state.shape[1:]
             self.renderer = Renderer(width, height, view_radius=2)
@@ -149,6 +149,10 @@ class SimpleFactory(BaseFactory):
                 else:
                     reward -= 0.5
 
+            else:
+                self.monitor.set('no_op', 1)
+                reward -= 0.25
+
             for entity in cols:
                 if entity != self.string_slices["dirt"]:
                     self.monitor.set(f'agent_{agent_state.i}_vs_{self.slice_strings[entity]}', 1)

main.py

@@ -49,27 +49,25 @@ def combine_runs(run_path: Union[str, PathLike]):
                                                          for col in df.columns if col not in ['episode', 'run']
                                                          }).reset_index()
 
+
     import seaborn as sns
     from matplotlib import pyplot as plt
-    df_melted = df_group.melt(id_vars=['train_step', 'run'],
+    df_melted = df_group.melt(id_vars=['episode', 'run'],
                               value_vars=['agent_0_vs_level', 'dirt_amount',
                                           'dirty_tiles', 'step_reward',
                                           'failed_cleanup_attempt',
                                           'dirt_cleaned'], var_name="Variable",
                               value_name="Score")
 
-    sns.lineplot(data=df_melted, x='train_step', y='Score', hue='Variable', ci='sd')
+    sns.lineplot(data=df_melted, x='episode', y='Score', hue='Variable', ci='sd')
     plt.show()
-
-    prepare_plot(filepath=run_path / f'{run_path.name}_monitor_out_combined',
-                 results_df=df.filter(regex=(".+_roll|(step)$")), tag='monitor')
     print('Plotting done.')
 
 
 if __name__ == '__main__':
 
-    # combine_runs('debug_out/PPO_1622113195')
+    combine_runs('debug_out/PPO_1622120377')
-    # exit()
+    exit()
 
     from stable_baselines3 import DQN, PPO
 
@@ -80,8 +78,9 @@ if __name__ == '__main__':
 
     for seed in range(5):
 
-        env = SimpleFactory(n_agents=1, dirt_properties=dirt_props)
+        env = SimpleFactory(n_agents=1, dirt_properties=dirt_props, force_skip_render=True)
-        model = PPO("MlpPolicy", env, verbose=1, ent_coef=0.0, seed=seed)
+
+        model = PPO("MlpPolicy", env, verbose=1, ent_coef=0.0, seed=seed, device='cpu')
 
         out_path = Path('../debug_out') / f'{model.__class__.__name__}_{time_stamp}'