Get columns from the entry

app.py

@@ -6,23 +6,13 @@ import pandas as pd
 
 from src.about import TITLE, INTRODUCTION_TEXT
 from src.display.css_html_js import custom_css
+from src.leaderboard.read_evals import TabularLeaderboardEntry
 from src.utils import download_github_file_content
 
 # GitHub API endpoint for the file (handles Git LFS files)
 LEADERBOARD_API_URL = "https://api.github.com/repos/upgini/mle-bench/contents/rankings/low/tabular/overall_ranks.csv"
 LEADERBOARD_GITHUB_URL = "https://github.com/upgini/mle-bench/blob/main/rankings/low/tabular/overall_ranks.csv"
 
-DISPLAY_COLUMNS = [
-    "experiment_id",
-    "Agent",
-    "LLM(s) used",
-    "mean_normalized_score",
-    "std_normalized_score",
-    "mean_medal_pct",
-    "sem_medal_pct",
-    "Date",
-]
-
 
 def download_leaderboard() -> pd.DataFrame:
     """Download the remote leaderboard CSV from GitHub (handles Git LFS).
@@ -32,14 +22,15 @@ def download_leaderboard() -> pd.DataFrame:
     csv_content = download_github_file_content(LEADERBOARD_API_URL, timeout=30)
 
     df = pd.read_csv(io.StringIO(csv_content))
+    display_columns = TabularLeaderboardEntry.get_display_columns()
     if df.empty:
-        return pd.DataFrame(columns=DISPLAY_COLUMNS)
+        return pd.DataFrame(columns=display_columns)
 
-    missing_cols = [col for col in DISPLAY_COLUMNS if col not in df.columns]
+    missing_cols = [col for col in display_columns if col not in df.columns]
     if missing_cols:
         raise ValueError(f"Leaderboard is missing expected columns: {', '.join(missing_cols)}")
 
-    df = df[DISPLAY_COLUMNS].copy()
+    df = df[display_columns].copy()
     df["mean_normalized_score"] = df["mean_normalized_score"].round(3)
     df["std_normalized_score"] = df["std_normalized_score"].round(3)
     df["mean_medal_pct"] = (df["mean_medal_pct"] * 100).round(1)
@@ -69,7 +60,7 @@ def create_app():
         gr.Markdown(INTRODUCTION_TEXT)
 
         leaderboard_table = gr.DataFrame(
-            value=pd.DataFrame(columns=DISPLAY_COLUMNS),
+            value=pd.DataFrame(columns=TabularLeaderboardEntry.get_display_columns()),
             wrap=True,
             interactive=False,
             type="pandas",

src/leaderboard/read_evals.py

@@ -21,6 +21,22 @@ class TabularLeaderboardEntry:
     sem_medal_pct: float
     date: str
 
+    @classmethod
+    def get_display_columns(cls) -> list[str]:
+        """Get the list of column names for DataFrame display."""
+        # Create a dummy instance to get the column order from to_dict()
+        dummy = cls(
+            experiment_id="",
+            agent="",
+            llms_used="",
+            mean_normalized_score=0.0,
+            std_normalized_score=0.0,
+            mean_medal_pct=0.0,
+            sem_medal_pct=0.0,
+            date="",
+        )
+        return list(dummy.to_dict().keys())
+
     @classmethod
     def from_dataframe_row(cls, row: pd.Series) -> "TabularLeaderboardEntry":
         """Create a TabularLeaderboardEntry from a pandas DataFrame row."""
@@ -69,18 +85,7 @@ def parse_tabular_leaderboard(df: pd.DataFrame) -> list[TabularLeaderboardEntry]
 def tabular_leaderboard_to_dataframe(entries: list[TabularLeaderboardEntry]) -> pd.DataFrame:
     """Convert a list of TabularLeaderboardEntry objects to a DataFrame."""
     if not entries:
-        return pd.DataFrame(
-            columns=[
-                "experiment_id",
-                "Agent",
-                "LLM(s) used",
-                "mean_normalized_score",
-                "std_normalized_score",
-                "mean_medal_pct",
-                "sem_medal_pct",
-                "Date",
-            ]
-        )
+        return pd.DataFrame(columns=TabularLeaderboardEntry.get_display_columns())
 
     data = [entry.to_dict() for entry in entries]
     return pd.DataFrame(data)

tests/test_leaderboard.py

@@ -4,7 +4,8 @@ import pandas as pd
 import pytest
 import requests
 
-from app import DISPLAY_COLUMNS, download_leaderboard, refresh_leaderboard
+from app import download_leaderboard, refresh_leaderboard
+from src.leaderboard.read_evals import TabularLeaderboardEntry
 
 
 @pytest.fixture
@@ -53,7 +54,7 @@ class TestDownloadLeaderboard:
         # Assertions
         assert isinstance(df, pd.DataFrame)
         assert len(df) == 3
-        assert list(df.columns) == DISPLAY_COLUMNS
+        assert list(df.columns) == TabularLeaderboardEntry.get_display_columns()
         mock_download.assert_called_once()
 
     @patch("app.download_github_file_content")
@@ -119,7 +120,7 @@ class TestDownloadLeaderboard:
 
         # Check that df is created correctly (extra columns should be filtered)
         assert len(df) == 2
-        assert list(df.columns) == DISPLAY_COLUMNS
+        assert list(df.columns) == TabularLeaderboardEntry.get_display_columns()
         # Verify the df doesn't have extra columns
         assert "extra_col" not in df.columns
 
@@ -167,14 +168,14 @@ class TestDownloadLeaderboard:
     @patch("app.download_github_file_content")
     def test_empty_dataframe(self, mock_download):
         """Test handling of empty CSV (header only)."""
-        csv_data = ",".join(DISPLAY_COLUMNS)  # Header only
+        csv_data = ",".join(TabularLeaderboardEntry.get_display_columns())  # Header only
         mock_download.return_value = csv_data
 
         df = download_leaderboard()
 
         assert isinstance(df, pd.DataFrame)
         assert len(df) == 0
-        assert list(df.columns) == DISPLAY_COLUMNS
+        assert list(df.columns) == TabularLeaderboardEntry.get_display_columns()
 
     @patch("app.download_github_file_content")
     def test_invalid_date_handling(self, mock_download):
@@ -207,7 +208,7 @@ class TestDownloadLeaderboard:
         # Should successfully download via download_url
         assert isinstance(df, pd.DataFrame)
         assert len(df) == 3
-        assert list(df.columns) == DISPLAY_COLUMNS
+        assert list(df.columns) == TabularLeaderboardEntry.get_display_columns()
         mock_download.assert_called_once()
 
     @patch("app.download_github_file_content")
@@ -220,7 +221,7 @@ class TestDownloadLeaderboard:
 
         assert isinstance(df, pd.DataFrame)
         assert len(df) == 3
-        assert list(df.columns) == DISPLAY_COLUMNS
+        assert list(df.columns) == TabularLeaderboardEntry.get_display_columns()
         mock_download.assert_called_once()