Upload 2 files

app.py

@@ -18,15 +18,14 @@ def load_data_and_model():
         return None, None
     df = pd.read_csv(df_path)
     
-    # Standardize ALL possible market value names
-    mv_rename_map = {}
-    for col in df.columns:
-        if 'market_value' in col.lower():
-            mv_rename_map[col] = 'market_value_in_eur'
-    if mv_rename_map:
-        df.rename(columns=mv_rename_map, inplace=True)
+    # Standardize ALL possible market value names to match model
+    # We look for ANY column containing 'market_value' and 'eur'
+    mv_cols = [c for c in df.columns if 'market_value' in c.lower()]
+    if mv_cols:
+        target_col = next((c for c in mv_cols if 'eur' in c.lower()), mv_cols[0])
+        df.rename(columns={target_col: 'market_value_in_eur'}, inplace=True)
     
-    # Deduplicate
+    # Deduplicate early
     df = df.loc[:, ~df.columns.duplicated()].copy()
     
     # Load model
@@ -37,13 +36,9 @@ def load_data_and_model():
             path = "FairValue_xgboost.json"
             
         if os.path.exists(path):
-            # Check file size to catch corrupted uploads
-            filesize = os.path.getsize(path)
-            if filesize < 100:
-                return df, f"CORRUPTED_MODEL (Size: {filesize} bytes. Is too small!)"
             model.load_model(path)
         else:
-            return df, f"MODEL_FILE_NOT_FOUND (Checked root for {path})"
+            return df, "MODEL_FILE_NOT_FOUND"
     except Exception as e:
         return df, f"LOAD_ERROR: {str(e)}"
         
@@ -70,10 +65,16 @@ if input_mode == "Select Existing Player":
         selected_name = st.sidebar.selectbox("Target Database Player", player_list)
         player_data = df[df[name_col] == selected_name].iloc[0:1].copy()
     else:
+        st.sidebar.error("Player name column missing.")
         st.stop()
+        
+    st.sidebar.markdown("---")
+    st.sidebar.markdown("**Simulation Parameters**")
     
+    # Use position-based access (iloc) for everything to avoid KeyError
     contract_default = max(0.5, min(6.0, float(player_data['Contract_Years_Left'].iloc[0])))
     age_default = max(16, min(40, int(player_data['Age'].iloc[0])))
+    
     injury_col = next((c for c in ['Injury_Days_Total_24m', 'Injury_Days'] if c in player_data.columns), None)
     injury_default = 0
     if injury_col:
@@ -84,7 +85,10 @@ if input_mode == "Select Existing Player":
     injuries = st.sidebar.number_input("Injury missed days (24m)", 0, 500, injury_default)
     
 else:
+    st.sidebar.markdown("---")
+    st.sidebar.markdown("**Custom Player Attributes**")
     player_data = df.median(numeric_only=True).to_frame().T
+    
     contract_years = st.sidebar.slider("Contract Years Remaining", 0.5, 6.0, 2.0, 0.5)
     age = st.sidebar.slider("Age", 16, 40, 24)
     injuries = st.sidebar.number_input("Injury missed days (24m)", 0, 500, 10)
@@ -95,32 +99,37 @@ else:
 
 asking_price = st.sidebar.number_input("Selling Club Asking Price (£m)", 1.0, 300.0, 45.0)
 
-# Build X_infer
-expected_cols = model.feature_names_in_
+# Build X_infer with safety
 player_data['Contract_Years_Left'] = contract_years
 player_data['Age'] = age
 if 'Injury_Days_Total_24m' in player_data.columns:
     player_data['Injury_Days_Total_24m'] = injuries
 
+expected_cols = model.feature_names_in_
 X_infer = player_data.reindex(columns=expected_cols, fill_value=0)
 
 if st.button("Calculate Prediction", type="primary"):
-    # Debugging Visibility
-    raw_preds = model.predict(X_infer)
-    log_pv = raw_preds[0]
-    baseline_pv = np.expm1(log_pv)
-
-    # UI logic
-    conservative_bound = baseline_pv * 0.85 
-    risk_discount_pct = 0.0
-    if contract_years < 1.5: risk_discount_pct += 0.20
-    if age > 30: risk_discount_pct += 0.15
-    if injuries > 60: risk_discount_pct += 0.10
+    # SAFE CHECK FOR ANY KEYERRORS
+    mv_v = 0.0
+    if 'market_value_in_eur' in X_infer.columns:
+        mv_v = X_infer['market_value_in_eur'].iloc[0]
+    
+    if mv_v == 0:
+        st.warning("⚠️ Market Value is 0.0. Prediction may be inaccurate.")
 
-    discount_val = conservative_bound * risk_discount_pct
-    hard_cap = conservative_bound - discount_val
+    log_pv = model.predict(X_infer)[0]
+    baseline_pv = np.expm1(log_pv)
 
+    # UI Logic
     baseline_pv_m = baseline_pv / 1_000_000
+    conservative_bound = baseline_pv * 0.85
+    
+    risk_discount = 0.0
+    if contract_years < 1.5: risk_discount += 0.20
+    if age > 30: risk_discount += 0.15
+    if injuries > 60: risk_discount += 0.10
+    
+    hard_cap = conservative_bound * (1 - risk_discount)
     hard_cap_m = hard_cap / 1_000_000
 
     col1, col2 = st.columns([2, 1])
@@ -137,27 +146,12 @@ if st.button("Calculate Prediction", type="primary"):
 
     with col2:
         st.subheader("Metrics Breakdown")
-        st.metric("Predicted Market Value (ML Baseline)", f"£{baseline_pv_m:.1f}m")
+        st.metric("Predicted Market Value (Baseline)", f"£{baseline_pv_m:.1f}m")
         st.metric("Recommended Hard Cap", f"£{hard_cap_m:.1f}m")
         if asking_price > hard_cap_m:
-            st.error(f"Winner's Curse Risk: £{asking_price - hard_cap_m:.1f}m Overpay.")
-
-    with st.expander("🔬 Technical Diagnostics (FOR DEBUGGING)"):
-        st.write("Raw Log-Prediction from Model:", log_pv)
-        st.write("Input Feature Vector (X_infer):", X_infer)
-        if log_pv == 0 and X_infer['market_value_in_eur'].iloc[0] > 0:
-            st.error("CRITICAL: The model is receiving data but returning exactly 0.0. This means the uploaded model file is likely empty or corrupted.")
-
-    st.subheader("Algorithmic Interpretability (SHAP)")
-    try:
-        explainer = shap.TreeExplainer(model)
-        shap_values = explainer(X_infer)
-        shap_df = pd.DataFrame({'Feature': X_infer.columns, 'Impact': shap_values.values[0]}).sort_values(by='Impact', ascending=False)
-        st.dataframe(shap_df.head(10), use_container_width=True)
-    except:
-        st.warning("SHAP Error. Model file/version mismatch.")
+            st.error(f"Winner's Curse Risk: £{asking_price-hard_cap_m:.1f}m Overpay")
 
 st.markdown("---")
 st.subheader("Model Performance Assessment")
-st.markdown(f"**Training Set Size:** `{len(df):,}` transfers | **Features:** `{len(expected_cols)}` cols")
+st.markdown(f"**Training Set:** `{len(df):,}` transfers | **Features:** `{len(expected_cols)}` cols")
 st.markdown("**Validation MAE:** `~£23,980,000` | **Engine:** XGBoost + SHAP")