updated xgboost metrics

app.py

@@ -223,7 +223,7 @@ with open("complete_results_acad.json", 'r') as f:
 # Recompute Elo scores for the full public dataset
 public_per_dataset = compute_elo_for_subset(public_per_dataset, metric="Accuracy")
 # Compute percentage improvement over XGBoost
-public_per_dataset = compute_pct_improvement_over_baseline(public_per_dataset, baseline_model="xgboost_ensemble", metric="Accuracy")
+public_per_dataset = compute_pct_improvement_over_baseline(public_per_dataset, baseline_model="xgboost", metric="Accuracy")
 public_enter_per_dataset = public_per_dataset
 
 with open("complete_results_indus.json", "r") as f:
@@ -232,7 +232,7 @@ with open("complete_results_indus.json", "r") as f:
 # Recompute Elo scores for the full private dataset
 private_per_dataset = compute_elo_for_subset(private_per_dataset, metric="Accuracy")
 # Compute percentage improvement over XGBoost
-private_per_dataset = compute_pct_improvement_over_baseline(private_per_dataset, baseline_model="xgboost_ensemble", metric="Accuracy")
+private_per_dataset = compute_pct_improvement_over_baseline(private_per_dataset, baseline_model="xgboost", metric="Accuracy")
 
 with open("qrt.json", "r") as f:
     qrt_scores = json.load(f)
@@ -449,14 +449,14 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
                 gr.Markdown('  ')
                 with gr.Row(equal_height=False):
                     with gr.Column(scale=8):
-                        public_model_agg = public_per_dataset.groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', 'Pct_Improvement_over_XGBoost']].mean().reset_index()
-                        public_enter_model_agg = public_enter_per_dataset.groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', 'Pct_Improvement_over_XGBoost']].mean().reset_index()
+                        public_model_agg = public_per_dataset.groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', '%↗ over XGBoost']].mean().reset_index()
+                        public_enter_model_agg = public_enter_per_dataset.groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', '%↗ over XGBoost']].mean().reset_index()
                         gr.Markdown("## 🏆 Overview of the Leaderboard *(Evaluation on publicly available Enterprise Datasets)*")
                         gr.Markdown('  ')
 
                         with gr.Row():
                             metric_selector = gr.Dropdown(
-                                choices=["Accuracy", "Precision", "Recall", "F1_score", "AUC", "Elo_score", "Pct_Improvement_over_XGBoost"],
+                                choices=["Accuracy", "Precision", "Recall", "F1_score", "AUC", "Elo_score", "%↗ over XGBoost"],
                                 value="Accuracy",
                                 label="📊 Metric to display",
                                 elem_classes=["compact-dropdown"]
@@ -543,7 +543,7 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
             gr.Markdown("## 🏆 Overview ")
 
             # Compute per-model averages from public_per_dataset
-            public_model_agg = public_per_dataset.groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', 'Pct_Improvement_over_XGBoost']].mean().reset_index()
+            public_model_agg = public_per_dataset.groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', '%↗ over XGBoost']].mean().reset_index()
             public_model_agg = public_model_agg.round(3)
             # Add model_type from model_info.json
             with open("model_info.json", "r") as f:
@@ -551,7 +551,7 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
 
             public_model_agg = public_model_agg.merge(model_info[["model_name", "model_type"]], left_on="model", right_on="model_name", how="left").drop(columns=["model_name"])
             # Only keep the required columns, with model_type second
-            public_model_agg = public_model_agg[['model', 'model_type', 'Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', 'Pct_Improvement_over_XGBoost']]
+            public_model_agg = public_model_agg[['model', 'model_type', 'Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', '%↗ over XGBoost']]
             public_model_agg = public_model_agg.sort_values(by=["model"], key=lambda x: x != "NICL")
             model_df = gr.DataFrame(
                 value=highlight_max(public_model_agg),
@@ -630,7 +630,7 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
                         "F1_score",
                         "AUC",
                         "Elo_score",
-                        "Pct_Improvement_over_XGBoost",
+                        "%↗ over XGBoost",
                     ]
 
                     # --- Filtering & aggregation ---
@@ -650,7 +650,7 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
                         # Recompute Elo scores for this filtered subset
                         filtered = compute_elo_for_subset(filtered, metric="Accuracy")
                         # Recompute percentage improvement over XGBoost
-                        filtered = compute_pct_improvement_over_baseline(filtered, baseline_model="xgboost_ensemble", metric="Accuracy")
+                        filtered = compute_pct_improvement_over_baseline(filtered, baseline_model="xgboost", metric="Accuracy")
 
                         per_model_avg = (
                             filtered
@@ -701,11 +701,11 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
                 # Recompute Elo scores for this industry subset
                 domain_per_dataset = compute_elo_for_subset(domain_per_dataset, metric="Accuracy")
                 # Recompute percentage improvement over XGBoost for this industry subset
-                domain_per_dataset = compute_pct_improvement_over_baseline(domain_per_dataset, baseline_model="xgboost_ensemble", metric="Accuracy")
+                domain_per_dataset = compute_pct_improvement_over_baseline(domain_per_dataset, baseline_model="xgboost", metric="Accuracy")
 
                 model_agg = (
                     domain_per_dataset
-                    .groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', 'Pct_Improvement_over_XGBoost']]
+                    .groupby('model')[['Accuracy', 'Precision', 'Recall', 'F1_score', 'AUC', 'Elo_score', '%↗ over XGBoost']]
                     .mean()
                     .round(3)
                     .reset_index()
@@ -762,7 +762,7 @@ with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
                     "F1_score",
                     "AUC",
                     "Elo_score",
-                    "Pct_Improvement_over_XGBoost",
+                    "%↗ over XGBoost",
                 ]   
                 
                 # Available datasets

utils.py

@@ -10,7 +10,7 @@ import scipy
 
 def compute_pct_improvement_over_baseline(
     df: pd.DataFrame,
-    baseline_model: str = "xgboost_ensemble",
+    baseline_model: str = "xgboost",
     metric: str = "Accuracy"
 ) -> pd.DataFrame:
     """
@@ -45,7 +45,7 @@ def compute_pct_improvement_over_baseline(
             return None
         return ((row[metric] - baseline) / baseline) * 100
 
-    df["Pct_Improvement_over_XGBoost"] = df.apply(calc_pct_improvement, axis=1)
+    df["%↗ over XGBoost"] = df.apply(calc_pct_improvement, axis=1)
 
     return df