app.py

import gradio as gr
import pandas as pd
import numpy as np
import joblib

# -----------------------------
# Config
# -----------------------------
THRESHOLD = 0.35  # tuned threshold
MODEL_PATH = "ensemble_model.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'
]

# -----------------------------
# Load model
# -----------------------------
model = joblib.load("main/ensemble_model_DEP_Class.pkl")

# -----------------------------
# 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):
    """
    Take human-friendly month number and other features, return model-ready DataFrame.
    """
    # Compute cyclical month encoding
    month_sin = np.sin(2 * np.pi * month / 12)
    month_cos = np.cos(2 * np.pi * month / 12)

    # 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
    }

    return pd.DataFrame([[input_dict[f] for f in FEATURES]], columns=FEATURES)

# -----------------------------
# 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
    )

    # Predict probability
    proba = model.predict_proba(X_ready)[:, 1][0]

    # Apply threshold
    pred = int(proba >= THRESHOLD)

    return {
        "Predicted_Class": "Delay" if pred == 1 else "On-time",
        "Probability_of_Delay": round(float(proba), 3)
    }

# -----------------------------
# 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")
]

outputs = gr.JSON(label="Prediction Result")

demo = gr.Interface(
    fn=predict_delay_gradio,
    inputs=inputs,
    outputs=outputs,
    title="Flight Delay Predictor",
    description="Enter simple month number and other features to predict flight delay probability."
)

if __name__ == "__main__":
    demo.launch()