#!/usr/bin/env python3
# Author: Simon-Pierre Boucher — contact@spboucher.ai
#
"""Step 3 — Fit hedonic models A-E and export the paper's result tables.

Inputs : data/processed/hedonic_maison_results.csv
Outputs: results/model_comparison.csv        (Table 4 of the paper)
         results/coefficients_model_D.csv    (full model, Table 5/6 source)
         results/coefficients_model_E.csv    (parsimonious model)
         results/descriptive_stats.csv       (Table 2 source)
         results/similarity_stats.csv        (Table 3 source)
"""

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.models import fit_all_models
from src.references import ENGLISH_LABELS, SIM_COLS

MODEL_NAMES = {
    "A": "Structural only",
    "B": "+ Description length",
    "C": "+ Semantic similarities (20)",
    "D": "Full model (B + C)",
    "E": "Parsimonious (significant only)",
}


def coefficient_table(model, variables):
    """Coefficients with HC3 errors, p-values and percentage price impacts."""
    rows = []
    for v in variables:
        if v not in model.params:
            continue
        coef = model.params[v]
        rows.append({
            "variable": v,
            "label": ENGLISH_LABELS.get(v.replace("sim_", ""), v),
            "coefficient": coef,
            "std_error": model.bse[v],
            "t_value": model.tvalues[v],
            "p_value": model.pvalues[v],
            "impact_pct": (np.exp(coef) - 1) * 100,
        })
    return pd.DataFrame(rows)


def main():
    df = pd.read_csv(config.ANALYSIS_CSV)
    print(f"{len(df):,} observations")

    models, specs, sig_sims = fit_all_models(df)
    config.RESULTS_DIR.mkdir(parents=True, exist_ok=True)

    comparison = pd.DataFrame({
        "model": list(models),
        "specification": [MODEL_NAMES[k] for k in models],
        "r2": [m.rsquared for m in models.values()],
        "adj_r2": [m.rsquared_adj for m in models.values()],
        "aic": [m.aic for m in models.values()],
        "bic": [m.bic for m in models.values()],
        "k": [int(m.df_model) for m in models.values()],
        "n": [int(m.nobs) for m in models.values()],
    })
    comparison["delta_r2_vs_A"] = comparison["r2"] - models["A"].rsquared
    comparison.to_csv(config.RESULTS_DIR / "model_comparison.csv", index=False)
    print(comparison.round(4).to_string(index=False))

    # F-test of the semantic block (D vs A), as reported in the paper
    mA, mD = models["A"], models["D"]
    df_diff = mD.df_model - mA.df_model
    f_stat = ((mA.ssr - mD.ssr) / df_diff) / (mD.ssr / mD.df_resid)
    from scipy import stats
    f_pval = 1 - stats.f.cdf(f_stat, df_diff, mD.df_resid)
    print(f"\nJoint F-test (D vs A): F = {f_stat:.2f}, p = {f_pval:.2e}")
    print(f"Significant similarities in D (p<0.05): {len(sig_sims)}/{len(SIM_COLS)}")

    coefficient_table(mD, specs["D"]).to_csv(
        config.RESULTS_DIR / "coefficients_model_D.csv", index=False)
    coefficient_table(models["E"], specs["E"]).to_csv(
        config.RESULTS_DIR / "coefficients_model_E.csv", index=False)

    desc_vars = ["price", "log_price"] + config.STRUCTURAL_VARS + ["remarks_length"]
    df[desc_vars].describe().T.to_csv(config.RESULTS_DIR / "descriptive_stats.csv")

    sim_stats = df[SIM_COLS].describe().T
    sim_stats.index = [ENGLISH_LABELS[c.replace("sim_", "")] for c in sim_stats.index]
    sim_stats["corr_log_price"] = [
        df[c].corr(df["log_price"]) for c in SIM_COLS
    ]
    sim_stats.to_csv(config.RESULTS_DIR / "similarity_stats.csv")

    print(f"\n-> {config.RESULTS_DIR}/model_comparison.csv, coefficients_model_D.csv, "
          f"coefficients_model_E.csv, descriptive_stats.csv, similarity_stats.csv")


if __name__ == "__main__":
    main()