#!/usr/bin/env python3
# Author: Simon-Pierre Boucher — contact@spboucher.ai
#
"""Step 2 — Embed descriptions and compute the 20 cosine-similarity features.

Inputs : data/processed/houses.parquet
Outputs: data/processed/embeddings_maisons.npy   (17,087 x 384, cached)
         data/processed/sim_matrix_maisons.npy    (17,087 x 20)
         data/processed/hedonic_maison_results.csv (analysis dataset)

Pass --force to re-encode the embeddings even if a valid cache exists.
"""

import sys
from pathlib import Path

import numpy as np
import pandas as pd

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))

from src import config
from src.embeddings import (add_similarity_columns, encode_references,
                            load_or_encode_remarks, similarity_features)
from src.references import SIM_COLS


def main(force=False):
    df = pd.read_parquet(config.HOUSES_PARQUET)
    print(f"{len(df):,} houses loaded")

    print("Encoding 20 reference descriptions ...")
    _, ref_embeddings = encode_references()

    print("Encoding listing descriptions (cache: data/processed) ...")
    prop_embeddings, from_cache = load_or_encode_remarks(
        df["remarks"], force=force
    )
    print(f"  embeddings {prop_embeddings.shape} "
          f"({'loaded from cache' if from_cache else 'freshly encoded'})")

    sim_matrix = similarity_features(prop_embeddings, ref_embeddings)
    np.save(config.SIM_MATRIX_NPY, sim_matrix)
    add_similarity_columns(df, sim_matrix)

    export_cols = ["id", "price", "log_price", "bedrooms", "bathrooms",
                   "half_baths", "parking", "stories", "land_size",
                   "remarks_length"] + SIM_COLS
    df[export_cols].to_csv(config.ANALYSIS_CSV, index=False)
    print(f"  -> {config.SIM_MATRIX_NPY}")
    print(f"  -> {config.ANALYSIS_CSV}")


if __name__ == "__main__":
    main(force="--force" in sys.argv)