Update app.py

app.py

@@ -31,19 +31,46 @@ async def get_results():
     try:
         # Load the dataset
         dataset = load_dataset("smolagents/results")
-        
         # Convert to list for processing
-        data = dataset["train"].to_pandas()
+        df = dataset["train"].to_pandas()
         
         # Log some info to help debug
-        print("Dataset loaded, shape:", data.shape)
-        print("Columns:", data.columns)
-        print("First row:", data.iloc[0])
-        
-        # Process the data to group by model and calculate scores
-        processed_data = []
-        grouped = data.groupby('model_id')
+        print("Dataset loaded, shape:", df.shape)
+        print("Columns:", df.columns)
+
+        # Process the data to match frontend expectations
+        result = []
+        # Ensure we have the expected columns
+        expected_columns = ['model_id', 'agent_action_type', 'benchmark', 'score']
+        for col in expected_columns:
+            if col not in df.columns:
+                print(f"Warning: Column {col} not found in dataset")
+                
+        # Group by model_id and agent_action_type to create the expected structure
+        for (model_id, agent_action_type), group in df.groupby(['model_id', 'agent_action_type']):
+            # Calculate scores for each benchmark
+            benchmark_scores = {}
+            benchmarks = ['GAIA', 'MATH', 'SimpleQA']
+            
+            for benchmark in benchmarks:
+                benchmark_group = group[group['benchmark'] == benchmark]
+                if not benchmark_group.empty:
+                    benchmark_scores[benchmark] = benchmark_group['score'].mean() * 100  # Convert to percentage
+            
+            # Calculate average if we have at least one benchmark score
+            if benchmark_scores:
+                benchmark_scores['Average'] = sum(benchmark_scores.values()) / len(benchmark_scores)
+            
+            # Add entry to result
+            result.append({
+                'model_id': model_id,
+                'agent_action_type': agent_action_type,
+                'scores': benchmark_scores
+            })
         
+        print(f"Processed {len(result)} entries for the frontend")
+        # Return the properly formatted data as a JSON response
+        return result 
         return data
         
     except Exception as e: