import os
import json
import joblib
import pandas as pd
import streamlit as st
from huggingface_hub import hf_hub_download

st.set_page_config(page_title="Wellness Package Predictor", page_icon="🏝️", layout="centered")
st.title("🏝️ Wellness Tourism Package — Purchase Propensity")

def get_secret(name: str, default=None):
    try:
        return st.secrets[name]
    except Exception:
        return os.getenv(name, default)

MODEL_REPO = get_secret("MODEL_REPO", "MBG0903/tourism_customer_xgb")
HF_TOKEN   = get_secret("HF_TOKEN", None)

@st.cache_resource(show_spinner=True)
def load_artifacts(repo_id: str, token: str | None):
    # hf_hub_download automatically finds the correct writable cache directory
    model_path = hf_hub_download(repo_id=repo_id, filename="model.joblib",
                                 token=token)
    meta_path  = hf_hub_download(repo_id=repo_id, filename="metadata.json",
                                 token=token)
    model = joblib.load(model_path)
    with open(meta_path, "r") as f:
        meta = json.load(f)
    return model, meta

try:
    model, meta = load_artifacts(MODEL_REPO, HF_TOKEN)
except Exception as e:
    st.error(f"Failed to load model artifacts from {MODEL_REPO}. Details: {e}")
    st.stop()

st.caption("Model metrics (from training)")
st.json(meta.get("metrics", {}))

# ---- UI ----
st.sidebar.header("Enter Customer Profile")
def i_num(label, value, minv=None, maxv=None, step=1):
    return st.sidebar.number_input(label, value=value, min_value=minv, max_value=maxv, step=step)

inputs = {}
inputs["CustomerID"] = st.sidebar.text_input("CustomerID", "CUST_000001")
inputs["Age"] = i_num("Age", 32, 18, 90)
inputs["TypeofContact"] = st.sidebar.selectbox("TypeofContact", ["Company Invited","Self Inquiry"])
inputs["CityTier"] = st.sidebar.selectbox("CityTier", ["Tier 1","Tier 2","Tier 3"])
inputs["Occupation"] = st.sidebar.selectbox("Occupation", ["Salaried","Freelancer","Self Employed","Student","Retired"])
inputs["Gender"] = st.sidebar.selectbox("Gender", ["Male","Female"])
inputs["NumberOfPersonVisiting"] = i_num("NumberOfPersonVisiting", 2, 1, 10)
inputs["PreferredPropertyStar"] = i_num("PreferredPropertyStar", 4, 1, 5)
inputs["MaritalStatus"] = st.sidebar.selectbox("MaritalStatus", ["Single","Married","Divorced"])
inputs["NumberOfTrips"] = i_num("NumberOfTrips", 3, 0, 50)
inputs["Passport"] = st.sidebar.selectbox("Passport", [0,1])
inputs["OwnCar"] = st.sidebar.selectbox("OwnCar", [0,1])
inputs["NumberOfChildrenVisiting"] = i_num("NumberOfChildrenVisiting", 0, 0, 10)
inputs["Designation"] = st.sidebar.selectbox("Designation", ["Executive","Manager","Senior Manager","AVP","VP","Director"])
inputs["MonthlyIncome"] = i_num("MonthlyIncome", 70000, 0, 1_000_000, 1000)
inputs["PitchSatisfactionScore"] = i_num("PitchSatisfactionScore", 4, 1, 5)
inputs["ProductPitched"] = st.sidebar.selectbox("ProductPitched", ["Basic","Deluxe","Super Deluxe","King","Queen"])
inputs["NumberOfFollowups"] = i_num("NumberOfFollowups", 2, 0, 20)
inputs["DurationOfPitch"] = i_num("DurationOfPitch", 15, 0, 120)

df_in = pd.DataFrame([inputs])
for junk in ["Unnamed: 0", "index"]:
    if junk in df_in.columns:
        df_in = df_in.drop(columns=[junk])

feature_order = meta.get("feature_order")
if feature_order:
    missing = [c for c in feature_order if c not in df_in.columns]
    if missing:
        st.warning(f"Input missing expected columns: {missing}")
    df_in = df_in[[c for c in feature_order if c in df_in.columns]]

threshold = float(meta.get("threshold", 0.5))
if st.button("Predict"):
    try:
        proba = float(model.predict_proba(df_in)[:, 1][0])
        pred = int(proba >= threshold)
        st.metric("Purchase Probability", f"{proba:.3f}")
        st.write("Prediction:", "Will Purchase (1)" if pred else "Will Not Purchase (0)")
        with st.expander("Input snapshot"):
            st.dataframe(df_in)
    except Exception as e:
        st.error(f"Prediction failed: {e}")