Update app.py

app.py

@@ -1,72 +1,119 @@
 import gradio as gr
-import joblib
-import numpy as np
 import pandas as pd
+import numpy as np
+import joblib
+
+# -----------------------------
+# Config
+# -----------------------------
+THRESHOLD = 0.35  # tuned threshold
+MODEL_PATH = "ensemble_model.pkl"
 
-# ===== Load Saved Models =====
-# Classification ensemble
-clf1, clf2, clf3, scaler_clf, weights_clf = joblib.load("main/ensemble_model_class.pkl")
-feature_names_clf = joblib.load("main/feature_names_class.pkl")
+FEATURES = [
+    'month_sin', 'month_cos', 'dep_dayofweek', 'is_weekend',
+    'distance_bin_ord', 'season', 'part_of_day_ord',
+    'wind_speed_cat_ord', 'humidity_cat_ord', 'precip_cat_ord',
+    'pressure_cat_ord', 'late_aircraft_flag', 'is_congested',
+    'origin_3day_delay_rate', 'carrier_3day_delay_rate'
+]
 
-# Regression ensemble
-linear_model_reg, ridge_model_reg, rf_model_reg, scaler_reg, weights_reg = joblib.load("main/ensemble_model.pkl")
-feature_names_reg = joblib.load("main/feature_names_regression.pkl")
+# -----------------------------
+# Load model
+# -----------------------------
+model = joblib.load("main/ensemble_model_DEP_Class.pkl")
 
-# ===== Prediction Function =====
-def classify_then_regress(*args):
+# -----------------------------
+# Feature preparation
+# -----------------------------
+def prepare_features_from_simple_month(month, dep_dayofweek, is_weekend,
+                                       distance_bin_ord, season, part_of_day_ord,
+                                       wind_speed_cat_ord, humidity_cat_ord, precip_cat_ord,
+                                       pressure_cat_ord, late_aircraft_flag, is_congested,
+                                       origin_3day_delay_rate, carrier_3day_delay_rate):
     """
-    Runs classification first; if class != 0, runs regression.
-    Uses separate feature sets for classification and regression.
+    Take human-friendly month number and other features, return model-ready DataFrame.
     """
-    # --- Classification step ---
-    input_dict_clf = dict(zip(feature_names_clf, args[:len(feature_names_clf)]))
-    df_clf = pd.DataFrame([input_dict_clf])
-
-    X_scaled_clf = scaler_clf.transform(df_clf)
-    pred1 = clf1.predict_proba(X_scaled_clf)
-    pred2 = clf2.predict_proba(X_scaled_clf)
-    pred3 = clf3.predict_proba(df_clf)  # RF-like model doesn't need scaling
+    # Compute cyclical month encoding
+    month_sin = np.sin(2 * np.pi * month / 12)
+    month_cos = np.cos(2 * np.pi * month / 12)
 
-    ensemble_proba = (pred1 * weights_clf[0] +
-                      pred2 * weights_clf[1] +
-                      pred3 * weights_clf[2])
+    # Build dict in correct order
+    input_dict = {
+        'month_sin': month_sin,
+        'month_cos': month_cos,
+        'dep_dayofweek': dep_dayofweek,
+        'is_weekend': is_weekend,
+        'distance_bin_ord': distance_bin_ord,
+        'season': season,
+        'part_of_day_ord': part_of_day_ord,
+        'wind_speed_cat_ord': wind_speed_cat_ord,
+        'humidity_cat_ord': humidity_cat_ord,
+        'precip_cat_ord': precip_cat_ord,
+        'pressure_cat_ord': pressure_cat_ord,
+        'late_aircraft_flag': late_aircraft_flag,
+        'is_congested': is_congested,
+        'origin_3day_delay_rate': origin_3day_delay_rate,
+        'carrier_3day_delay_rate': carrier_3day_delay_rate
+    }
 
-    class_idx = int(np.argmax(ensemble_proba, axis=1)[0])
-    confidence = float(ensemble_proba[0, class_idx])
+    return pd.DataFrame([[input_dict[f] for f in FEATURES]], columns=FEATURES)
 
-    # If classification result is 0 → stop here
-    if class_idx == 0:
-        return f"Class {class_idx} (Confidence: {confidence:.2%})", "Skipped — class is 0"
+# -----------------------------
+# Prediction function
+# -----------------------------
+def predict_delay_gradio(month, dep_dayofweek, is_weekend,
+                         distance_bin_ord, season, part_of_day_ord,
+                         wind_speed_cat_ord, humidity_cat_ord, precip_cat_ord,
+                         pressure_cat_ord, late_aircraft_flag, is_congested,
+                         origin_3day_delay_rate, carrier_3day_delay_rate):
+    # Prepare features
+    X_ready = prepare_features_from_simple_month(
+        month, dep_dayofweek, is_weekend,
+        distance_bin_ord, season, part_of_day_ord,
+        wind_speed_cat_ord, humidity_cat_ord, precip_cat_ord,
+        pressure_cat_ord, late_aircraft_flag, is_congested,
+        origin_3day_delay_rate, carrier_3day_delay_rate
+    )
 
-    # --- Regression step ---
-    input_dict_reg = dict(zip(feature_names_reg, args[:len(feature_names_reg)]))
-    df_reg = pd.DataFrame([input_dict_reg])
+    # Predict probability
+    proba = model.predict_proba(X_ready)[:, 1][0]
 
-    X_scaled_reg = scaler_reg.transform(df_reg)
-    pred_linear = linear_model_reg.predict(X_scaled_reg)
-    pred_ridge = ridge_model_reg.predict(X_scaled_reg)
-    pred_rf = rf_model_reg.predict(df_reg)
+    # Apply threshold
+    pred = int(proba >= THRESHOLD)
 
-    ensemble_pred = (pred_linear * weights_reg[0] +
-                     pred_ridge * weights_reg[1] +
-                     pred_rf * weights_reg[2])
+    return {
+        "Predicted_Class": "Delay" if pred == 1 else "On-time",
+        "Probability_of_Delay": round(float(proba), 3)
+    }
 
-    return f"Class {class_idx} (Confidence: {confidence:.2%})", float(ensemble_pred)
+# -----------------------------
+# Gradio UI
+# -----------------------------
+inputs = [
+    gr.Number(label="Month (1-12)"),
+    gr.Number(label="dep_dayofweek"),
+    gr.Number(label="is_weekend"),
+    gr.Number(label="distance_bin_ord"),
+    gr.Number(label="season"),
+    gr.Number(label="part_of_day_ord"),
+    gr.Number(label="wind_speed_cat_ord"),
+    gr.Number(label="humidity_cat_ord"),
+    gr.Number(label="precip_cat_ord"),
+    gr.Number(label="pressure_cat_ord"),
+    gr.Number(label="late_aircraft_flag"),
+    gr.Number(label="is_congested"),
+    gr.Number(label="origin_3day_delay_rate"),
+    gr.Number(label="carrier_3day_delay_rate")
+]
 
-# ===== Build Gradio Inputs =====
-# Use the union of both feature sets so all required inputs are present
-all_features = list(dict.fromkeys(feature_names_clf + feature_names_reg))
-inputs = [gr.Number(label=col) for col in all_features]
+outputs = gr.JSON(label="Prediction Result")
 
 demo = gr.Interface(
-    fn=classify_then_regress,
+    fn=predict_delay_gradio,
     inputs=inputs,
-    outputs=[
-        gr.Textbox(label="Classification Result"),
-        gr.Textbox(label="Regression Prediction")
-    ],
-    title="Conditional Classification → Regression",
-    description="Runs classification first; if class != 0, runs regression."
+    outputs=outputs,
+    title="Flight Delay Predictor",
+    description="Enter simple month number and other features to predict flight delay probability."
 )
 
 if __name__ == "__main__":