Update the text based

CustomDataset/Structured/BA2Motif.py

@@ -9,7 +9,7 @@ class BA2Motif(Base):
     A class to process the BA2Motif dataset, find frequent patterns, and visualize graphs and patterns.
     """
 
-    def __init__(self, path='../rawData/BA2Motif'):
+    def __init__(self, path='../rawData'):
         """
         Initialize the BA2Motif processor.
 

CustomDataset/Structured/MultiShape.py

@@ -9,7 +9,7 @@ class MultiShape(Base):
     A class to process the BAMultiShapes dataset, find frequent patterns, and visualize graphs and patterns.
     """
 
-    def __init__(self, path='../rawData/BAMultiShapes'):
+    def __init__(self, path='../rawData'):
         """
         Initialize the MultiShape processor.
 

GNN/HeterogenousGNN.py

@@ -123,7 +123,7 @@ class GNN(torch.nn.Module):
 
     def train_model(self, epochs: int = 75, lr: float = 0.0005,
                     train_split: float = 0.8, show_progress: bool = False,
-                    seed: Optional[int] = 8) -> Dict[str, Dict[str, float]]:
+                    seed: Optional[int] = 15) -> Dict[str, Dict[str, float]]:
         
         # Set random seeds for reproducibility
         torch.manual_seed(seed)

GNN/HomogeneousGNN.py

@@ -91,10 +91,10 @@ class GNN(torch.nn.Module):
         
         return F.log_softmax(x, dim=1)
 
-    def train_model(self, epochs: int = 300, lr: float = 0.001, 
+    def train_model(self, epochs: int = 75, lr: float = 0.0005, 
                 train_split: float = 0.8, show_progress: bool = False) -> Dict[str, float]:
         """
-        Train with enhanced learning schedule and early stopping
+        Train with enhanced learning schedule but without early stopping
         """
         train_loader, test_loader = self._prepare_data(train_split)
 
@@ -103,8 +103,6 @@ class GNN(torch.nn.Module):
                                     patience=20, min_lr=1e-5)
 
         best_metrics = {'accuracy': 0, 'precision': 0, 'recall': 0, 'f1': 0, 'epoch': 0}
-        patience = 150
-        no_improve = 0
         
         # Enhanced class weight calculation
         if train_loader is not None:
@@ -175,9 +173,6 @@ class GNN(torch.nn.Module):
             if current_metrics['f1'] > best_metrics['f1']:
                 best_metrics = {**current_metrics, 'epoch': epoch}
                 self.best_model_state = self.state_dict()
-                no_improve = 0
-            else:
-                no_improve += 1
 
             if show_progress and (epoch + 1) % 10 == 0:
                 print(f"\nEpoch {epoch+1:03d}, Loss: {total_loss/len(train_loader):.4f}")
@@ -189,11 +184,6 @@ class GNN(torch.nn.Module):
                     for metric, value in current_metrics.items():
                         print(f"Test {metric.capitalize()}: {value:.4f}")
 
-            # Enhanced early stopping
-            if no_improve >= patience:
-                print(f"\nEarly stopping triggered after {epoch + 1} epochs")
-                break
-        
         if self.best_model_state:
             self.load_state_dict(self.best_model_state)
         return best_metrics

Utils/Utils.py

@@ -96,7 +96,7 @@ def group_keywords_into_themes(keywords: List[str], num_themes: int):
     openai.api_key = api_key
     
     # Construct the prompt
-    prompt = f"Group the following keywords into {num_themes} distinct themes. Ensure each theme has a meaningful name, and do not combine themes unnecessarily. Here are the keywords:\n\n{', '.join(keywords)}\n\nPlease return the themes in a JSON format where the key is the theme name and the value is a list of grouped keywords."
+    prompt = f"Group the following keywords into {num_themes} distinct themes. Ensure each theme has a meaningful name, and do not combine themes unnecessarily. Uuse all the keywords while grouping into themes. Here are the keywords:\n\n{', '.join(keywords)}\n\nPlease return the themes in a JSON format where the key is the theme name and the value is a list of grouped keywords."
 
     # Call the OpenAI API with the correct method
     response = openai.ChatCompletion.create(

main.py

@@ -16,19 +16,20 @@ def parse_arguments():
     parser.add_argument("-d", "--dataset", help="Specific dataset name to run, optional")
     parser.add_argument("-n", "--num_groups", type=int, nargs='*', default=None, help="List of group sizes to run experiments with, space-separated")
     parser.add_argument("-l", "--labels", type=int, nargs='*', default=None, help="List of labels to run experiments with, space-separated")
-    parser.add_argument("-b", "--boolean_concepts", type=str, choices=['true', 'false'], default=None, help="Whether to create high level concepts as boolean values, must explicitly state 'true' or 'false'")
+    parser.add_argument("-b", "--boolean_concepts", type=str, choices=['true', 'false'], default=None, help="Whether to create high-level concepts as boolean values, must explicitly state 'true' or 'false'")
+    parser.add_argument("-e", "--use_experimented_groups", action='store_true', help="Use experimented groups if set, otherwise create new grouped keywords")
+    parser.add_argument("-p", "--penalty", type=int, default=1, help="Set the penalty of evolearner")
     parser.add_argument("--title", type=str, default="", help="Title to append to the results folder name")
     args = parser.parse_args()
 
-    # Convert boolean_concepts argument from string to actual Boolean type, or None if not specified
     if args.boolean_concepts is not None:
         args.boolean_concepts = True if args.boolean_concepts.lower() == 'true' else False
+    else:
+        args.boolean_concepts = True
 
-    # Check if the iterations argument is a positive integer
     if args.iterations < 1:
         parser.error("The number of iterations must be a positive integer.")
 
-    # Clean title: remove spaces and special characters, convert to lowercase
     if args.title:
         args.title = "_" + "".join(c.lower() for c in args.title if c.isalnum())
 
@@ -48,7 +49,7 @@ def main():
         types_to_run = ['text', 'structured']
     
     for dataset_type in types_to_run:
-        datasets = config[dataset_type]  # Load datasets based on type (structured or text)
+        datasets = config[dataset_type]
         
         # If a specific dataset is specified, filter to only run that one
         if args.dataset:
@@ -65,18 +66,20 @@ def main():
                     iterations=args.iterations,
                     create_high_level_concepts_as_boolean=args.boolean_concepts,
                     selected_labels=args.labels,
-                    title=args.title
+                    title=args.title,
+                    penalty=args.penalty
                 )
             elif dataset_type == "text":
                 text_based_datasets_experiment(
-                    dataset["grouped_keyword_dir"],
                     dataset["dataset_name"],
                     dataset["entity_name"],
                     iterations=args.iterations,
                     num_groups_list=num_groups_list,
                     create_high_level_concepts_as_boolean=args.boolean_concepts,
                     selected_labels=args.labels,
-                    title=args.title 
+                    title=args.title,
+                    use_experimented_groups=args.use_experimented_groups,
+                    penalty=args.penalty
                 )
 
 if __name__ == "__main__":

requirements.txt

@@ -16,7 +16,7 @@ networkx==3.3
 nltk==3.9.1
 numpy==2.1.0
 ontolearn==0.6.0
-openai==1.59.7
+openai==0.27.0
 psutil==6.0.0
 pyparsing==3.1.2
 requests==2.32.3
@@ -25,8 +25,9 @@ scipy==1.14.0
 sympy==1.13.2
 threadpoolctl==3.5.0
 torch==2.4.0
-torch_geometric==2.5.3
+torch_geometric==2.6.1
 tqdm==4.66.5
 urllib3==2.2.2
 yarl==1.9.4
 python-dotenv==1.0.1
+pytest==8.3.3
\ No newline at end of file

tests/test_owl_file_gereation.py

@@ -88,7 +88,7 @@ def test_generate_owl_file(custom_heterodata):
 def test_discriminative_result_dataproperties(custom_heterodata):
         hgnn = GNN
         owl_graph_path = os.path.join(base_dir, "testGraphs/test_discriminative_result_dataproperties.owl")  # No leading slash
-        explainer = DiscriminativeExplainer(hgnn, custom_heterodata, "http://example.org/", owl_graph_path, generate_new_owl_file=True)
+        explainer = DiscriminativeExplainer(hgnn, custom_heterodata, "http://example.org/", owl_graph_path, generate_new_owl_file=False)
         hypotheses, model = explainer.explain(1, 1, use_data_properties=True, debug=True)
         hypothesis = hypotheses[0]
         assert hypothesis.quality == 1.0
@@ -100,7 +100,7 @@ def test_discriminative_result_dataproperties(custom_heterodata):
 def test_discriminative_result_objectproperties(custom_heterodata1):
         hgnn = GNN
         owl_graph_path = os.path.join(base_dir, "testGraphs/test_discriminative_result_objectproperties.owl")  # No leading slash
-        explainer = DiscriminativeExplainer(hgnn, custom_heterodata1, "http://example.org/", owl_graph_path, generate_new_owl_file=True)
+        explainer = DiscriminativeExplainer(hgnn, custom_heterodata1, "http://example.org/", owl_graph_path, generate_new_owl_file=False)
         hypotheses, model = explainer.explain(1, 1, use_data_properties=False, debug=True)
         hypothesis = hypotheses[0]
         assert hypothesis.quality == 1.0

text_based_datasets_experiment.py

@@ -33,7 +33,7 @@ def run_gnn(structuredDataset: Base, entity_name, datasetName, results_dir):
     print("Initializing GNN model...")
     model = GNN(structuredDataset.dataset)
     print("Training model...")
-    metrics = model.train_model(epochs=150, lr=0.0005, show_progress=True)
+    metrics = model.train_model()
     evaluations["gnn"] = metrics[entity_name]
 
     print("\nGNN Metrics:")
@@ -112,7 +112,7 @@ def append_to_csv_file(results, results_dir, dataset_key, num_groups, create_hig
 
     print(f"Results appended to {filename}")
 
-def explain_and_evaluate(model, dataset, entity_name, owl_graph_path, high_level_concepts, create_high_level_concepts_as_boolean, selected_labels=None):
+def explain_and_evaluate(model, dataset, entity_name, owl_graph_path, high_level_concepts, create_high_level_concepts_as_boolean, selected_labels=None, penalty=1):
     """
     Explains and evaluates each label in the dataset.
     """
@@ -134,7 +134,7 @@ def explain_and_evaluate(model, dataset, entity_name, owl_graph_path, high_level
         label_name = dataset[entity_name].yLabel[label]
         start_time = time.time()  # Start timing for each explanation
         print(f"\nExplaining {entity_name} label {label}: {label_name}")
-        hypotheses, model = explainer.explain(label, 5, max_runtime=90, num_generations=1000)
+        hypotheses, model = explainer.explain(label, 5, max_runtime=90, num_generations=1000, penalty=penalty)
         explain_time = time.time() - start_time  # End timing for each explanation
         
         best_hypothesis = hypotheses[0].concept
@@ -215,7 +215,7 @@ def summarize_aggregated_results(aggregated_results, results_dir, dataset_name):
 
     print(f"Summary results saved to {summary_filename}")
 
-def experiment(grouped_keyword_dir, dataset_name, entity_name, iterations=5, num_groups_list=[0, 5, 10, 15, 20, 25], create_high_level_concepts_as_boolean=True, selected_labels=None, title=""):
+def experiment(dataset_name, entity_name, iterations=5, num_groups_list=[0, 5, 10, 15, 20, 25], create_high_level_concepts_as_boolean=True, selected_labels=None, title="", use_experimented_groups=False, penalty=1):
     """
     Handles dataset loading and evaluation for experiments.
     Manages the experiment based on specified number of groups and boolean concept creation settings.
@@ -225,11 +225,20 @@ def experiment(grouped_keyword_dir, dataset_name, entity_name, iterations=5, num
 
     # Create timestamp directory for all results
     timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
-    
     results_dir = Path(f"evaluation_results/{timestamp}_{dataset_name}{title}")
-
     results_dir.mkdir(parents=True, exist_ok=True)
 
+    # Create directory for grouped keywords
+    if use_experimented_groups:
+        grouped_keyword_dir = f"experimentedGroups/{dataset_name}"
+    else:
+        grouped_keyword_dir = results_dir / "groupedKeywords"
+        grouped_keyword_dir.mkdir(parents=True, exist_ok=True)
+
+    # Load dataset and train GNN model only once
+    dataset = load_datasets(dataset_name=dataset_name)
+    model = run_gnn(dataset, entity_name, dataset_name, results_dir)
+
     aggregated_results = {}
 
     for run in range(1, iterations + 1):
@@ -238,9 +247,6 @@ def experiment(grouped_keyword_dir, dataset_name, entity_name, iterations=5, num
         # Create run-specific directory
         run_dir = create_directory(results_dir, f"run_{run}")
 
-        dataset = load_datasets(dataset_name=dataset_name)
-        model = run_gnn(dataset, entity_name, dataset_name, run_dir)
-
         write_header = True
         for num_groups in num_groups_list:
             group_keyword_file = "" if num_groups == 0 else os.path.join(grouped_keyword_dir, f'groupedKeywords_{num_groups}.json')
@@ -253,7 +259,7 @@ def experiment(grouped_keyword_dir, dataset_name, entity_name, iterations=5, num
             high_level_concepts = None if num_groups == 0 else fetch_high_level_concepts(dataset, num_groups, group_keyword_file)
 
             results = explain_and_evaluate(
-                model, dataset.dataset, entity_name, owl_graph_path, high_level_concepts, create_high_level_concepts_as_boolean, selected_labels
+                model, dataset.dataset, entity_name, owl_graph_path, high_level_concepts, create_high_level_concepts_as_boolean, selected_labels, penalty
             )
 
             append_to_csv_file(results, run_dir, dataset_name, num_groups, create_high_level_concepts_as_boolean, write_header=write_header)
@@ -291,24 +297,22 @@ def experiment(grouped_keyword_dir, dataset_name, entity_name, iterations=5, num
 def main():
     datasets = [
         {
-            'grouped_keyword_dir': 'rawData/dblp/groups',
             'dataset_name': 'dblp',
             'entity_name': 'author'
         }
-        # ,
-        # {
-        #     'grouped_keyword_dir': 'rawData/imdb/groups',
-        #     'dataset_name': 'imdb',
-        #     'entity_name': 'movie'
-        # }
+        ,
+        {
+            'dataset_name': 'imdb',
+            'entity_name': 'movie'
+        }
     ]
 
     for dataset in datasets:
         experiment(
-            dataset["grouped_keyword_dir"],
             dataset["dataset_name"],
             dataset["entity_name"],
-            iterations=5
+            iterations=5,
+            use_experimented_groups=True
         )
 
 if __name__ == "__main__":