debug

app.py

@@ -54,25 +54,28 @@ from dataclasses import fields
 
 class SmartSelectColumns(SelectColumns):
     """
-    Enhanced SelectColumns component for gradio_leaderboard with explicit column grouping.
+    Enhanced SelectColumns component for gradio_leaderboard with dynamic column filtering.
     """
     def __init__(
         self,
-        column_groups: Optional[Dict[str, List[str]]] = None,
+        benchmark_keywords: Optional[List[str]] = None,
+        model_keywords: Optional[List[str]] = None,
         column_mapping: Optional[Dict[str, str]] = None,
         initial_selected: Optional[List[str]] = None,
         **kwargs
     ):
         """
-        Initialize SmartSelectColumns with enhanced functionality.
+        Initialize SmartSelectColumns with dynamic filtering.
         
         Args:
-            column_groups: Dict mapping group names to lists of columns in that group
+            benchmark_keywords: List of benchmark names to filter by (e.g., ["ioi", "mcqa"])
+            model_keywords: List of model names to filter by (e.g., ["llama3", "qwen2_5"])
             column_mapping: Dict mapping actual column names to display names
             initial_selected: List of columns to show initially
         """
         super().__init__(**kwargs)
-        self.column_groups = column_groups or {}
+        self.benchmark_keywords = benchmark_keywords or []
+        self.model_keywords = model_keywords or []
         self.column_mapping = column_mapping or {}
         self.reverse_mapping = {v: k for k, v in self.column_mapping.items()} if column_mapping else {}
         self.initial_selected = initial_selected or []
@@ -85,6 +88,40 @@ class SmartSelectColumns(SelectColumns):
         """Transform actual column names to display names."""
         return [self.column_mapping.get(col, col) for col in y]
 
+    def get_filtered_groups(self, df: pd.DataFrame) -> Dict[str, List[str]]:
+        """
+        Dynamically create column groups based on keywords.
+        """
+        filtered_groups = {}
+        
+        # Create benchmark groups
+        for benchmark in self.benchmark_keywords:
+            matching_cols = [
+                col for col in df.columns 
+                if benchmark in col.lower()
+            ]
+            if matching_cols:
+                group_name = f"Benchmark group for {benchmark}"
+                filtered_groups[group_name] = [
+                    self.column_mapping.get(col, col) 
+                    for col in matching_cols
+                ]
+        
+        # Create model groups
+        for model in self.model_keywords:
+            matching_cols = [
+                col for col in df.columns 
+                if model in col.lower()
+            ]
+            if matching_cols:
+                group_name = f"Model group for {model}"
+                filtered_groups[group_name] = [
+                    self.column_mapping.get(col, col) 
+                    for col in matching_cols
+                ]
+        
+        return filtered_groups
+
     def update(
         self, 
         value: Union[pd.DataFrame, Dict[str, List[str]], Any]
@@ -97,14 +134,8 @@ class SmartSelectColumns(SelectColumns):
             # Use initial selection or default columns
             selected = self.initial_selected if self.initial_selected else choices
             
-            # Convert column groups to use display names
-            filtered_cols = {}
-            for group_name, columns in self.column_groups.items():
-                filtered_cols[group_name] = [
-                    self.column_mapping.get(col, col) 
-                    for col in columns 
-                    if col in value.columns
-                ]
+            # Get dynamically filtered groups
+            filtered_cols = self.get_filtered_groups(value)
             
             return {
                 "choices": choices,
@@ -134,7 +165,6 @@ class SmartSelectColumns(SelectColumns):
 
 
 
-
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
@@ -348,52 +378,24 @@ def init_leaderboard_mib_subgraph(dataframe, track):
     #     interactive=False,
     # )
         # Complete column groups for both benchmarks and models
-    column_groups = {
-        # Benchmark groups
-        "Benchmark group for ioi": ["ioi_gpt2", "ioi_qwen2_5", "ioi_gemma2", "ioi_llama3"],
-        "Benchmark group for mcqa": ["mcqa_qwen2_5", "mcqa_gemma2", "mcqa_llama3"],
-        "Benchmark group for arithmetic_addition": ["arithmetic_addition_llama3"],
-        "Benchmark group for arithmetic_subtraction": ["arithmetic_subtraction_llama3"],
-        "Benchmark group for arc_easy": ["arc_easy_gemma2", "arc_easy_llama3"],
-        "Benchmark group for arc_challenge": ["arc_challenge_llama3"],
-        
-        # Model groups
-        "Model group for qwen2_5": ["ioi_qwen2_5", "mcqa_qwen2_5"],
-        "Model group for gpt2": ["ioi_gpt2"],
-        "Model group for gemma2": ["ioi_gemma2", "mcqa_gemma2", "arc_easy_gemma2"],
-        "Model group for llama3": [
-            "ioi_llama3", 
-            "mcqa_llama3", 
-            "arithmetic_addition_llama3", 
-            "arithmetic_subtraction_llama3", 
-            "arc_easy_llama3", 
-            "arc_challenge_llama3"
-        ]
-    }
+    # Define keywords for filtering
+    benchmark_keywords = ["ioi", "mcqa", "arithmetic_addition", "arithmetic_subtraction", "arc_easy", "arc_challenge"]
+    model_keywords = ["qwen2_5", "gpt2", "gemma2", "llama3"]
 
-    # # Complete mappings for more readable display names
+    # # Optional: Define display names
     # mappings = {
-    #     # IOI benchmark mappings
     #     "ioi_llama3": "IOI (LLaMA-3)",
     #     "ioi_qwen2_5": "IOI (Qwen-2.5)",
     #     "ioi_gpt2": "IOI (GPT-2)",
     #     "ioi_gemma2": "IOI (Gemma-2)",
-        
-    #     # MCQA benchmark mappings
     #     "mcqa_llama3": "MCQA (LLaMA-3)",
     #     "mcqa_qwen2_5": "MCQA (Qwen-2.5)",
     #     "mcqa_gemma2": "MCQA (Gemma-2)",
-        
-    #     # Arithmetic benchmark mappings
     #     "arithmetic_addition_llama3": "Arithmetic Addition (LLaMA-3)",
     #     "arithmetic_subtraction_llama3": "Arithmetic Subtraction (LLaMA-3)",
-        
-    #     # ARC benchmark mappings
     #     "arc_easy_llama3": "ARC Easy (LLaMA-3)",
     #     "arc_easy_gemma2": "ARC Easy (Gemma-2)",
     #     "arc_challenge_llama3": "ARC Challenge (LLaMA-3)",
-        
-    #     # Other columns
     #     "eval_name": "Evaluation Name",
     #     "Method": "Method",
     #     "Average": "Average Score"
@@ -402,12 +404,13 @@ def init_leaderboard_mib_subgraph(dataframe, track):
 
     # Create SmartSelectColumns instance
     smart_columns = SmartSelectColumns(
-        column_groups=column_groups,
+        benchmark_keywords=benchmark_keywords,
+        model_keywords=model_keywords,
         column_mapping=mappings,
         initial_selected=["Method", "Average"]
     )
 
-    # Create Leaderboard directly
+    # Create Leaderboard
     leaderboard = Leaderboard(
         value=renamed_df,
         datatype=[c.type for c in fields(AutoEvalColumn_mib_subgraph)],
@@ -422,6 +425,8 @@ def init_leaderboard_mib_subgraph(dataframe, track):
 
 
 
+
+
 # def init_leaderboard_mib_subgraph(dataframe, track):
 #     """Initialize the subgraph leaderboard with group-based column selection."""
 #     if dataframe is None or dataframe.empty: