app.py

# app.py - UI mapped to your Business Context & Data Dictionary
import os
import json
import streamlit as st
import logging
from predict_utils import load_model, predict

# -------------------------
# Logging
# -------------------------
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("wellness_app")

# -------------------------
# Page config
# -------------------------
st.set_page_config(page_title="Visit with Us — Wellness Package Predictor", layout="centered")
st.title("Visit with Us — Wellness Package Purchase Predictor")
st.write(
    "Fill customer & interaction details below. Click **Predict** to estimate whether this customer is likely to purchase the Wellness Tourism Package."
)

# -------------------------
# Feature order used when calling the model
# If your training pipeline used different names/order, modify this list to match it exactly.
# Exclude CustomerID and ProdTaken (target).
FEATURE_ORDER = [
    "Age",
    "TypeofContact",
    "CityTier",
    "Occupation",
    "Gender",
    "NumberOfPersonVisiting",
    "PreferredPropertyStar",
    "MaritalStatus",
    "NumberOfTrips",
    "Passport",
    "OwnCar",
    "NumberOfChildrenVisiting",
    "Designation",
    "MonthlyIncome",
    "PitchSatisfactionScore",
    "ProductPitched",
    "NumberOfFollowups",
    "DurationOfPitch"
]

# -------------------------
# Helper conversions
# -------------------------
def to_int_or_none(v):
    try:
        if v is None or str(v).strip() == "":
            return None
        return int(v)
    except:
        try:
            return int(float(v))
        except:
            return None

def to_float_or_none(v):
    try:
        if v is None or str(v).strip() == "":
            return None
        return float(v)
    except:
        return None

def to_bool_int(v):
    # expects 0/1 or True/False or 'Yes'/'No'
    if v in (0, "0", "False", "false", False):
        return 0
    if v in (1, "1", "True", "true", True):
        return 1
    if isinstance(v, str):
        s = v.strip().lower()
        if s in ("yes", "y", "true", "t", "1"):
            return 1
        if s in ("no", "n", "false", "f", "0"):
            return 0
    return None

# -------------------------
# Debug / env info (collapsible)
# -------------------------
with st.expander("Environment & debug info", expanded=False):
    st.text(f"HF_MODEL_REPO = {os.getenv('HF_MODEL_REPO')}")
    st.text(f"HF_MODEL_FILENAME = {os.getenv('HF_MODEL_FILENAME')}")
    st.text(f"HF_TOKEN present? {bool(os.getenv('HF_TOKEN'))}")
    st.text(f"Working dir = {os.getcwd()}")
    try:
        st.write("Files in working directory:", sorted(os.listdir(".")))
    except Exception as e:
        st.write("Could not list directory:", e)

# -------------------------
# Load model once
# -------------------------
@st.cache_resource(show_spinner=False)
def _load():
    return load_model()

model = _load()
if model is None:
    st.error("Model not loaded. Check server logs for details. The UI remains available for testing.")
    st.stop()
st.success("Model loaded successfully ✅")

# -------------------------
# Input form (matching Data Dictionary)
# -------------------------
st.header("Customer details")
with st.form(key="customer_form"):
    # CustomerID intentionally optional (not used by model)
    customer_id = st.text_input("CustomerID (optional)", value="")

    age = st.number_input("Age", min_value=0, max_value=120, value=35, step=1, help="Age in years")

    typeof_contact = st.selectbox("Type of Contact", options=["Company Invited", "Self Inquiry"], index=0)
    city_tier = st.selectbox("City Tier", options=[1, 2, 3], index=1, help="1 = Tier 1 (most developed)")

    occupation = st.selectbox("Occupation", options=[
        "Salaried", "Freelancer", "Business", "Student", "Retired", "Other"
    ], index=0)

    gender = st.selectbox("Gender", options=["Male", "Female", "Other"], index=0)

    num_persons = st.number_input("Number Of Persons Visiting", min_value=1, max_value=20, value=1, step=1)

    preferred_property_star = st.selectbox("Preferred Property Star", options=[1, 2, 3, 4, 5], index=3)

    marital_status = st.selectbox("Marital Status", options=["Single", "Married", "Divorced", "Widowed"], index=1)

    number_of_trips = st.number_input("Average Number Of Trips per Year", min_value=0, max_value=100, value=2, step=1)

    passport = st.selectbox("Has Passport?", options=["No", "Yes"], index=1)
    own_car = st.selectbox("Owns Car?", options=["No", "Yes"], index=0)

    num_children = st.number_input("Number Of Children Visiting (age <5)", min_value=0, max_value=10, value=0, step=1)

    designation = st.text_input("Designation (job title)", value="")

    monthly_income = st.number_input("Monthly Income (gross)", min_value=0.0, value=30000.0, step=1000.0, format="%.2f")

    st.markdown("---")
    st.header("Interaction / Pitch details")

    pitch_score = st.slider("Pitch Satisfaction Score (1-10)", min_value=0.0, max_value=10.0, value=7.0, step=0.1)
    product_pitched = st.selectbox("Product Pitched", options=[
        "Wellness Package", "Adventure Package", "Luxury Package", "Budget Package", "Family Package", "Other"
    ], index=0)
    number_of_followups = st.number_input("Number Of Followups", min_value=0, max_value=100, value=1, step=1)
    duration_of_pitch = st.number_input("Duration Of Pitch (minutes)", min_value=0.0, max_value=1000.0, value=10.0, step=0.5)

    submit = st.form_submit_button("Predict")

# -------------------------
# Prepare payload & predict
# -------------------------
if submit:
    # prepare a mapping preserving FEATURE_ORDER
    payload = {
        "Age": to_int_or_none(age),
        "TypeofContact": typeof_contact,
        "CityTier": to_int_or_none(city_tier),
        "Occupation": occupation,
        "Gender": gender,
        "NumberOfPersonVisiting": to_int_or_none(num_persons),
        "PreferredPropertyStar": to_int_or_none(preferred_property_star),
        "MaritalStatus": marital_status,
        "NumberOfTrips": to_int_or_none(number_of_trips),
        "Passport": to_bool_int(passport),
        "OwnCar": to_bool_int(own_car),
        "NumberOfChildrenVisiting": to_int_or_none(num_children),
        "Designation": designation,
        "MonthlyIncome": to_float_or_none(monthly_income),
        "PitchSatisfactionScore": to_float_or_none(pitch_score),
        "ProductPitched": product_pitched,
        "NumberOfFollowups": to_int_or_none(number_of_followups),
        "DurationOfPitch": to_float_or_none(duration_of_pitch)
    }

    # Ensure final dict follows FEATURE_ORDER
    ordered_payload = {k: payload.get(k) for k in FEATURE_ORDER}

    st.markdown("### Payload sent to model (ordered)")
    st.json(ordered_payload)

    # call predict from predict_utils
    with st.spinner("Running prediction..."):
        result = predict(model, ordered_payload)

    if result is None:
        st.error("Prediction returned nothing. Check server logs.")
    elif "error" in result:
        st.error(f"Prediction error: {result['error']}")
    else:
        pred = result.get("prediction")
        prob = result.get("probability")
        st.success("Prediction complete")
        st.write("**Will Purchase?**", "Yes" if int(pred) == 1 else "No")
        if prob is not None:
            st.write("**Predicted probability:**", round(float(prob), 4))
        # helpful suggestion:
        if int(pred) == 1:
            st.info("Suggested action: prioritize outreach and personalize offer.")
        else:
            st.info("Suggested action: low priority; consider nurturing campaign.")

# -------------------------
# Quick utilities & debug
# -------------------------
st.markdown("---")
with st.expander("Quick test & debug", expanded=False):
    if st.button("Show model type & capabilities"):
        try:
            st.write("Model type:", type(model))
            st.write("Has predict:", hasattr(model, "predict"))
            st.write("Has predict_proba:", hasattr(model, "predict_proba"))
        except Exception as e:
            st.error(str(e))

    if st.button("Generate sample payload (example)"):
        sample = {
            "Age": 30,
            "TypeofContact": "Company Invited",
            "CityTier": 1,
            "Occupation": "Salaried",
            "Gender": "Female",
            "NumberOfPersonVisiting": 2,
            "PreferredPropertyStar": 4,
            "MaritalStatus": "Married",
            "NumberOfTrips": 3,
            "Passport": 1,
            "OwnCar": 0,
            "NumberOfChildrenVisiting": 1,
            "Designation": "Manager",
            "MonthlyIncome": 65000.0,
            "PitchSatisfactionScore": 8.5,
            "ProductPitched": "Wellness Package",
            "NumberOfFollowups": 2,
            "DurationOfPitch": 15.0
        }
        st.json(sample)
        st.write("You can paste this JSON into the 'Paste JSON dict' mode if present in earlier versions.")

st.caption("If your saved pipeline expects a different feature order or column names, update FEATURE_ORDER at the top of this file to match the training pipeline exactly.")