utils/scenario_engine.py

# utils/scenario_engine.py
import pandas as pd
import joblib
from utils.models import load_model  # your existing model loader

def simulate_scenario(feature, change, filter=None):
    """
    Dynamic scenario simulator that adapts to any dataset/model schema.
    """

    # 1️⃣ Load dataset and model
    df = pd.read_csv("appdata_rf.csv")
    model = load_model("models/app_best.joblib")

    # 2️⃣ Dynamically read model feature names
    expected_features = getattr(model, "feature_names_in_", list(df.select_dtypes(include='number').columns))
    print("Model expects:", expected_features)

    # 3️⃣ Parse change instruction
    try:
        pct_change = float(change.replace("%", "")) / 100 if "%" in change else float(change)
    except ValueError:
        return {"summary": f"⚠️ Invalid change value '{change}'", "df": df}

    # 4️⃣ Apply filter if any
    try:
        subset = df.query(filter) if filter else df.copy()
    except Exception as e:
        return {"summary": f"⚠️ Invalid filter '{filter}': {e}", "df": df}

    # 5️⃣ Apply the change dynamically
    if feature not in df.columns:
        return {"summary": f"⚠️ Feature '{feature}' not found in dataset.", "df": df}

    subset[feature] *= (1 + pct_change)
    if filter:
        df.loc[subset.index, feature] = subset[feature]
    else:
        df = subset

    # 6️⃣ Predict new churn rate using expected model inputs
    try:
        X = df[expected_features]
        churn_prob = model.predict_proba(X)[:, 1]
        new_churn_rate = churn_prob.mean() * 100
    except Exception as e:
        return {"summary": f"⚠️ Prediction error: {e}", "df": df}

    # 7️⃣ Return result
    summary = (
        f"📊 New churn rate = {new_churn_rate:.2f}% "
        f"after changing **{feature}** by {change}."
        + (f" (Filter: {filter})" if filter else "")
    )
    return {"summary": summary, "df": df}