Update app.py

app.py

@@ -1,40 +1,56 @@
 import gradio as gr
-import pandas as pd
+import pickle
 import numpy as np
 
-# Gerekli fonksiyonlar ve modeller buraya import edilmeli
-
-def predict_risk(cvss_score, cwe_id, description):
-    desc_len = len(description)
-    # Simülasyon (yerine gerçek model konmalı)
-    previous_pred = 4.5
-    
-    optimized = dynamic_risk_adjustment(previous_pred, cvss_score, cwe_id, desc_len)
-    confidence = calculate_confidence(0, cvss_score)
-    
-    if optimized <= 3.0:
-        category = "Düşük"
-    elif optimized <= 7.0:
-        category = "Orta"
-    else:
-        category = "Yüksek"
+# Modeli yükle
+with open("model/xgboost_optimized_v2.pkl", "rb") as f:
+    model_pkg = pickle.load(f)
+
+xgb_improved = model_pkg["base_models"]["xgboost_improved"][0]
+smart_align = model_pkg["enhancement_functions"]["smart_feature_alignment"]
+ensemble_fn = model_pkg["enhancement_functions"]["ensemble_predict"]
+unknown_boost_fn = model_pkg["enhancement_functions"]["unknown_category_boost"]
+dynamic_adjust_fn = model_pkg["enhancement_functions"]["dynamic_risk_adjustment"]
+confidence_fn = model_pkg["enhancement_functions"]["calculate_confidence"]
+
+preprocessors = model_pkg["preprocessors"]
+scaler = preprocessors["scaler"]
+encoders = preprocessors["encoders"]
+bert_extractor = preprocessors["bert_extractor"]
+label_encoder = preprocessors["label_encoder"]
+feature_names = preprocessors["scaler"].mean_.shape[0]
+
+def predict(cve_id, cwe_id, description, cvss_score):
+    # Dummy features for now – Hugging Face Spaces does not allow GPU by default
+    numeric = np.zeros((1, feature_names))
+    bert_feats = np.zeros((1, 768))  # Dummy
+    x_input = np.hstack([numeric, bert_feats])
+
+    xgb_pred = xgb_improved.predict(x_input)[0]
+    ensemble_pred = ensemble_fn(cvss_score, xgb_pred)
+    final_pred = unknown_boost_fn(ensemble_pred, unknown_features=2, cvss_score=cvss_score)
+    optimized = dynamic_adjust_fn(final_pred, cvss_score, cwe_id, len(description))
+    confidence = confidence_fn(2, cvss_score)
+
+    category = "Düşük" if optimized <= 3 else "Orta" if optimized <= 7 else "Yüksek"
 
     return {
-        "Risk Skoru": optimized,
+        "Tahmini Risk Skoru": round(optimized, 2),
         "Kategori": category,
-        "Güven": f"{confidence:.0%}"
+        "Model Güveni": f"%{round(confidence*100, 1)}"
     }
 
-iface = gr.Interface(
-    fn=predict_risk,
+interface = gr.Interface(
+    fn=predict,
     inputs=[
-        gr.Number(label="CVSS Skoru"),
-        gr.Text(label="CWE ID (örnek: CWE-89)"),
-        gr.Textbox(label="CVE Açıklaması", lines=5),
+        gr.Text(label="CVE ID"),
+        gr.Text(label="CWE ID"),
+        gr.Textbox(label="Açıklama"),
+        gr.Slider(minimum=0.0, maximum=10.0, step=0.1, label="CVSS Skoru")
     ],
     outputs="json",
-    title="Siber Risk Skoru Tahmini",
-    description="CVE/CWE verisinden risk tahmini yapan basit demo"
+    title="Risk Skoru Tahmin Modeli (XGBoost Optimized)",
+    description="CVE tanımlarına göre tahmini risk skoru ve kategori"
 )
 
-iface.launch()
+interface.launch()