debug

app.py

@@ -167,34 +167,85 @@ from src.about import TasksMib_Subgraph
 
 
 
+# def init_leaderboard_mib_subgraph(dataframe, track):
+#     if dataframe is None or dataframe.empty:
+#         raise ValueError("Leaderboard DataFrame is empty or None.")
+    
+#     # Add filter columns to display
+#     dataframe['Task'] = dataframe.apply(
+#         lambda row: [task.value.benchmark for task in TasksMib_Subgraph 
+#                     if any(f"{task.value.benchmark}_{model}" in row.index 
+#                           for model in task.value.models)][0], 
+#         axis=1
+#     )
+    
+#     dataframe['Model'] = dataframe.apply(
+#         lambda row: [model for task in TasksMib_Subgraph 
+#                     for model in task.value.models 
+#                     if f"{task.value.benchmark}_{model}" in row.index][0],
+#         axis=1
+#     )
+    
+#     return Leaderboard(
+#         value=dataframe,
+#         datatype=[c.type for c in fields(AutoEvalColumn_mib_subgraph)],
+#         select_columns=SelectColumns(
+#             default_selection=[c.name for c in fields(AutoEvalColumn_mib_subgraph) if c.displayed_by_default],
+#             cant_deselect=[c.name for c in fields(AutoEvalColumn_mib_subgraph) if c.never_hidden],
+#             label="Select Columns to Display:",
+#         ),
+#         search_columns=["Method", "Task", "Model"],  # Add Task and Model to searchable columns
+#         hide_columns=[c.name for c in fields(AutoEvalColumn_mib_subgraph) if c.hidden],
+#         bool_checkboxgroup_label="Hide models",
+#         interactive=False,
+#     )
+
 def init_leaderboard_mib_subgraph(dataframe, track):
+    """Initialize the subgraph leaderboard with grouped column selection."""
     if dataframe is None or dataframe.empty:
         raise ValueError("Leaderboard DataFrame is empty or None.")
+
+    # Get tasks and models using the new class methods
+    tasks = TasksMib_Subgraph.get_all_tasks()
+    models = TasksMib_Subgraph.get_all_models()
+
+    # Create a mapping from selection to actual column names
+    selection_map = {}
     
-    # Add filter columns to display
-    dataframe['Task'] = dataframe.apply(
-        lambda row: [task.value.benchmark for task in TasksMib_Subgraph 
-                    if any(f"{task.value.benchmark}_{model}" in row.index 
-                          for model in task.value.models)][0], 
-        axis=1
-    )
-    
-    dataframe['Model'] = dataframe.apply(
-        lambda row: [model for task in TasksMib_Subgraph 
-                    for model in task.value.models 
-                    if f"{task.value.benchmark}_{model}" in row.index][0],
-        axis=1
-    )
+    # Add task mappings - when a task is selected, show all its columns
+    for task in tasks:
+        # For each task, find all valid task_model combinations
+        valid_combos = []
+        for model in models:
+            col_name = f"{task}_{model}"
+            if col_name in dataframe.columns:
+                valid_combos.append(col_name)
+        if valid_combos:
+            selection_map[task] = valid_combos
     
+    # Add model mappings - when a model is selected, show all its columns
+    for model in models:
+        # For each model, find all valid task_model combinations
+        valid_combos = []
+        for task in tasks:
+            col_name = f"{task}_{model}"
+            if col_name in dataframe.columns:
+                valid_combos.append(col_name)
+        if valid_combos:
+            selection_map[model] = valid_combos
+
     return Leaderboard(
         value=dataframe,
         datatype=[c.type for c in fields(AutoEvalColumn_mib_subgraph)],
         select_columns=SelectColumns(
-            default_selection=[c.name for c in fields(AutoEvalColumn_mib_subgraph) if c.displayed_by_default],
-            cant_deselect=[c.name for c in fields(AutoEvalColumn_mib_subgraph) if c.never_hidden],
-            label="Select Columns to Display:",
+            choices=[tasks, models],  # Two groups of choices
+            labels=["Tasks", "Models"],  # Labels for each group
+            default_selection=[*tasks, *models],  # Show everything by default
+            cant_deselect=["Method"],  # Method column always visible
+            label="Filter by Tasks or Models:",
+            selection_map=selection_map  # Map selections to actual columns
         ),
-        search_columns=["Method", "Task", "Model"],  # Add Task and Model to searchable columns
+        search_columns=["Method"],
         hide_columns=[c.name for c in fields(AutoEvalColumn_mib_subgraph) if c.hidden],
         bool_checkboxgroup_label="Hide models",
         interactive=False,
@@ -211,6 +262,7 @@ def init_leaderboard_mib_subgraph(dataframe, track):
 
 
 
+
 def init_leaderboard_mib_causalgraph(dataframe, track):
     # print("Debugging column issues:")
     # print("\nActual DataFrame columns:")

old-requirements.txt

@@ -0,0 +1,17 @@
+APScheduler
+black
+datasets
+fastapi==0.112.2
+gradio
+gradio[oauth]
+gradio_leaderboard==0.0.13
+gradio_client
+huggingface-hub>=0.18.0
+matplotlib
+numpy
+pandas
+python-dateutil
+tqdm
+transformers
+tokenizers>=0.15.0
+sentencepiece

requirements.txt

@@ -4,7 +4,7 @@ datasets
 fastapi==0.112.2
 gradio
 gradio[oauth]
-gradio_leaderboard==0.0.13
+gradio_leaderboard==0.0.15
 gradio_client
 huggingface-hub>=0.18.0
 matplotlib

src/about.py

@@ -47,7 +47,18 @@ class TasksMib_Subgraph(Enum):
     task4 = TaskMIB_Subgraph("arc_easy", ["gemma2", "llama3"], "arc_easy", ["edge_counts", "faithfulness"])
     task5 = TaskMIB_Subgraph("arc_challenge", ["llama3"], "arc_challenge", ["edge_counts", "faithfulness"])
 
-
+    @classmethod
+    def get_all_tasks(cls):
+        """Returns a list of all task benchmarks"""
+        return [task.value.benchmark for task in cls]
+    
+    @classmethod
+    def get_all_models(cls):
+        """Returns a list of all unique models across all tasks"""
+        models = set()
+        for task in cls:
+            models.update(task.value.models)
+        return sorted(list(models))
 
 
 # @dataclass