Updates env, tests and utils

rl_core/environments/time_series.py

@@ -107,22 +107,23 @@ def __init__(self, primitives: list[str] = None, max_number_of_nodes: int = 10,
 
     def _get_obs(self) -> np.ndarray:
         """ Returns current environment's observation """
-        # graph_structure = self._get_graph_structure()
+        graph_structure = self._get_graph_structure()
 
-        node_structure = self._apply_one_hot_encoding(self._nodes_structure, self.number_of_primitives + 1)
-        edge_structure = self._edges_structure
+        # node_structure = self._apply_one_hot_encoding(self._nodes_structure, self.number_of_primitives + 1)
+        # edge_structure = self._edges_structure
 
-        # if self._meta_data is not None:
-        #     obs = np.concatenate((graph_structure, self._meta_data))
-        # else:
-        #     obs = graph_structure
-        #
+        if self._meta_data is not None:
+            obs = np.concatenate((graph_structure, self._meta_data))
+        else:
+            obs = graph_structure
 
-        obs = {
-            'meta': self._meta_data,
-            'nodes': node_structure,
-            'edges': edge_structure
-        }
+
+        # For sb3 models
+        # obs = {
+        #     'meta': self._meta_data,
+        #     'nodes': node_structure,
+        #     'edges': edge_structure
+        # }
 
         return obs
 
@@ -247,8 +248,18 @@ def step(self, action: int, mode: str = 'train') -> (np.ndarray, int, bool, bool
 
         assert action in self.action_space
 
+        # Checks if action is not valid
+        if not action in self._get_available_actions().keys():
+            terminated = False
+            truncated = False
+            reward = -10
+
+            self.env_step += 1
+            observation = self._get_obs()
+            info = self._get_info()
+
         # Checks if action is from special actions (e.g. eof - End of Pipeline)
-        if action in self._special_action.keys():
+        elif action in self._special_action.keys():
             terminated, truncated, reward = self._apply_eop_action()
 
             observation = self._get_obs()

rl_core/utils.py

@@ -4,7 +4,7 @@
 from sklearn.model_selection import train_test_split
 
 from meta_automl.utils import project_root
-from rl_core.dataloader import DataLoader_TS
+from rl_core.dataloader import TimeSeriesDataLoader
 
 OFFLINE_TRAJECTORIES = [
     [2, 0],
@@ -31,8 +31,8 @@ def define_data_for_experiment(test_size: int = 3):
     dataset_names = [name for name in os.listdir(data_folder_path)]
 
     temp = pd.read_csv('pipeline_validation_results.csv', index_col=0)
-    train = temp[temp['Topo Pipeline'].isna() == False]['Dataset'].to_list()
-    test = temp[temp['Topo Pipeline'].isna() == True]['Dataset'].to_list()
+    train = temp[temp['Topo Pipeline'].isna() == True]['Dataset'].to_list()
+    test = temp[temp['Topo Pipeline'].isna() == False]['Dataset'].to_list()
 
     # if test_size:
     #     train, test = train_test_split(dataset_names, test_size=3)
@@ -50,7 +50,7 @@ def define_data_for_experiment(test_size: int = 3):
     path_to_meta_data = os.path.join(str(project_root()),
                                      'MetaFEDOT\\data\\knowledge_base_time_series_0\\meta_features_ts.csv')
 
-    dataloader_train = DataLoader_TS(train_datasets, path_to_meta_data=path_to_meta_data)
-    dataloader_test = DataLoader_TS(test_datasets, path_to_meta_data=path_to_meta_data)
+    dataloader_train = TimeSeriesDataLoader(train_datasets, path_to_meta_data=path_to_meta_data)
+    dataloader_test = TimeSeriesDataLoader(test_datasets, path_to_meta_data=path_to_meta_data)
 
     return dataloader_train, dataloader_test, train, test,

tests/unit/rl_test/test_ts_env.py

@@ -71,7 +71,7 @@ def test_correct_pipelines(trajectory):
 def test_max_number_of_actions_in_pipelines(max_number_of_nodes):
     train_data, test_data = get_time_series()
 
-    env = TimeSeriesPipelineEnvironment(max_number_of_nodes=max_number_of_nodes, metadata_dim=0)
+    env = TimeSeriesPipelineEnvironment(max_number_of_nodes=max_number_of_nodes, metadata_dim=0, using_number_of_nodes=max_number_of_nodes)
     env.load_data(train_data, test_data, meta=None)
     env.reset()