Upload app.py with huggingface_hub

app.py

@@ -1,6 +1,6 @@
 """
-V38.2 Admission Model - Training Space (Residualized Version)
-Runs train_v38_2_resid.py on HF Space hardware and displays results via Gradio.
+V38.2-PRO Admission Model - Training Space
+Runs train_v38_2_pro.py with feature selection + optimized residualization.
 """
 import gradio as gr
 import subprocess
@@ -11,16 +11,14 @@ import time
 
 OUTPUT_DIR = './output'
 os.makedirs(OUTPUT_DIR, exist_ok=True)
-
-TRAIN_SCRIPT = 'train_v38_2_resid.py'
+TRAIN_SCRIPT = 'train_v38_2_pro.py'
 
 def run_training():
-    """Execute the V38.2 residualized model training"""
+    """Execute the V38.2-PRO model training"""
     log_lines = []
     log_lines.append("=" * 70)
-    log_lines.append(f"Starting V38.2 Residualized Model Training...")
+    log_lines.append(f"Starting V38.2-PRO Model Training...")
     log_lines.append(f"Script: {TRAIN_SCRIPT}")
-    log_lines.append(f"Working directory: {os.getcwd()}")
     log_lines.append("=" * 70)
     yield "\n".join(log_lines)
     
@@ -41,43 +39,32 @@ def run_training():
     process.wait()
     log_lines.append(f"\nProcess exited with code: {process.returncode}")
     
-    results_path = os.path.join(OUTPUT_DIR, 'v38_2_resid_results.json')
-    if os.path.exists(results_path):
-        with open(results_path) as f:
-            results = json.load(f)
-        log_lines.append("\n" + "=" * 70)
-        log_lines.append("RESULTS SUMMARY")
-        log_lines.append("=" * 70)
-        log_lines.append(json.dumps(results, indent=2))
+    for results_name in ['v38_2_pro_results.json', 'v38_2_resid_results.json']:
+        results_path = os.path.join(OUTPUT_DIR, results_name)
+        if os.path.exists(results_path):
+            with open(results_path) as f:
+                results = json.load(f)
+            log_lines.append("\n" + "=" * 70)
+            log_lines.append(f"RESULTS: {results_name}")
+            log_lines.append("=" * 70)
+            log_lines.append(json.dumps(results, indent=2))
+            break
     
     yield "\n".join(log_lines[-300:])
 
 def check_status():
-    results_path = os.path.join(OUTPUT_DIR, 'v38_2_resid_results.json')
-    if os.path.exists(results_path):
-        with open(results_path) as f:
-            results = json.load(f)
-        summary = f"""V38.2 Residualized Model Results:
-  Version: {results.get('version', 'N/A')}
-  Timestamp: {results.get('timestamp', 'N/A')}
-  Elapsed: {results.get('elapsed_minutes', 0):.1f} minutes
-  Features: {results.get('n_features', 'N/A')}
-  
-  GroupKFold AUC: {results['final_metrics']['auc']:.4f} (V37.3: 0.8697, V38.2-bare: 0.8687)
-  Temporal AUC: {results['temporal_validation']['avg_blend']:.4f} (V37.3: 0.8410, V38.2-bare: 0.8417)
-  
-  Top 10 Features:
-"""
-        for i, (fname, imp) in enumerate(results.get('feature_importance', [])[:10]):
-            summary += f"    {i+1}. {fname}: {imp:.2f}\n"
-        return summary
-    else:
-        return "No results yet. Click 'Run Training' to start."
+    for results_name in ['v38_2_pro_results.json', 'v38_2_resid_results.json']:
+        results_path = os.path.join(OUTPUT_DIR, results_name)
+        if os.path.exists(results_path):
+            with open(results_path) as f:
+                results = json.load(f)
+            return json.dumps(results, indent=2)
+    return "No results yet. Click Run Training to start."
 
 def auto_run():
-    results_path = os.path.join(OUTPUT_DIR, 'v38_2_resid_results.json')
+    results_path = os.path.join(OUTPUT_DIR, 'v38_2_pro_results.json')
     if not os.path.exists(results_path):
-        print("Auto-starting residualized model training...")
+        print("Auto-starting V38.2-PRO model training...")
         process = subprocess.Popen(
             [sys.executable, TRAIN_SCRIPT],
             stdout=subprocess.PIPE,
@@ -90,9 +77,9 @@ def auto_run():
         process.wait()
         print(f"Training complete. Exit code: {process.returncode}")
 
-with gr.Blocks(title="V38.2 Residualized Model") as demo:
-    gr.Markdown("# V38.2 Residualized Model - Simpson's Paradox Fix")
-    gr.Markdown("Adds school_base_rate, residualized features, explicit interactions, and school percentiles")
+with gr.Blocks(title="V38.2-PRO Model") as demo:
+    gr.Markdown("# V38.2-PRO Model - Feature Selection + Optimized Residualization")
+    gr.Markdown("Stage 1: Feature importance → Stage 2: Selected features + stronger regularization")
     
     with gr.Row():
         run_btn = gr.Button("Run Training", variant="primary")