import streamlit as st
from shared.hf_helpers import build_pipeline
from retailgpt_evaluator.leaderboard import build_leaderboard
import yaml, pandas as pd, os
from pathlib import Path

def main():
    # Safely set page config (won’t error inside streamlit_hub.py)
    try:
        st.set_page_config(page_title="RetailGPT Evaluator", page_icon="🛍️", layout="wide")
    except st.errors.StreamlitAPIException:
        pass

    st.title("🛍️ RetailGPT Evaluator — AxionX Digital")

    # Load config safely
    config_path = Path(__file__).resolve().parent / "config.yaml"
    with open(config_path) as f:
        cfg = yaml.safe_load(f)

    # Show leaderboard if exists
    leaderboard_df = None
    results_path = Path("models/retail_eval_results.json")
    if results_path.exists():
        try:
            leaderboard_df = build_leaderboard(results_path)
            st.subheader("📊 Model Leaderboard")
            st.dataframe(leaderboard_df, use_container_width=True)

            st.markdown("#### 📈 Evaluation Metrics")
            metric_options = leaderboard_df["model"].tolist()
            selected = st.selectbox("Inspect metrics for:", metric_options)
            selected_row = leaderboard_df[leaderboard_df["model"] == selected].iloc[0]

            cols = st.columns(4)
            cols[0].metric("ROUGE-L", f"{selected_row['rougeL']:.3f}")
            cols[1].metric("BLEU", f"{selected_row['bleu']:.3f}")
            cols[2].metric("Factuality", f"{selected_row['factuality']:.3f}")
            cols[3].metric("Score (avg)", f"{selected_row['score']:.3f}")

            st.bar_chart(
                leaderboard_df.set_index("model")[["rougeL", "bleu", "factuality", "score"]]
            )
        except Exception as exc:  # pragma: no cover - defensive UI fallback
            st.error(f"Unable to load evaluation results: {exc}")
    else:
        st.warning("Run `evaluate.py` first to generate metrics.")

    # Model chat interface
    st.markdown("---")
    model_name = st.selectbox("Choose a model to chat with:", cfg["models"])
    pipe = build_pipeline(model_name)

    query = st.text_area("Customer query:", "Which is the best country for retail?.")
    if st.button("Ask Model"):
        result = pipe(query, max_new_tokens=cfg["demo"]["max_new_tokens"])
        st.markdown("### 🧠 Model Response")
        st.write(result[0]["generated_text"])

# Ensure it's import-safe
if __name__ == "__main__":
    main()