"""Registered tests (plan/analysis-plan.md).
Usage: python code/analyze.py <bids_dir> <features.csv> <out.json>"""
import json
import sys

import numpy as np
import pandas as pd
import statsmodels.api as sm
from scipy import stats

SEED = 20261009
NBOOT = 5000


def rank(x):
    return stats.rankdata(x)


def rank_reg(d, y, x, covs, one_sided_positive=True):
    d = d.dropna(subset=[y, x] + covs)
    X = pd.DataFrame({c: rank(d[c]) if d[c].nunique() > 2 else d[c].astype(float) for c in [x] + covs})
    X = sm.add_constant(X)
    fit = sm.OLS(rank(d[y]), X).fit(cov_type="HC3")
    t, p2 = float(fit.tvalues[x]), float(fit.pvalues[x])
    # standardised coefficient on ranks
    beta = float(fit.params[x] * np.std(X[x]) / np.std(rank(d[y])))
    return {"t": round(t, 3), "beta_std": round(beta, 4), "p_two_sided": p2,
            "p_one_sided": float(stats.norm.sf(t)) if one_sided_positive else float(stats.norm.cdf(t)), "n": int(len(d))}


def holm(ps):
    order = sorted(ps, key=ps.get); out, run = {}, 0.0
    for i, k in enumerate(order):
        run = max(run, min(1.0, (len(ps) - i) * ps[k])); out[k] = round(run, 5)
    return out


def main():
    bids, feat, out = sys.argv[1:4]
    q = pd.read_csv(f"{bids}/phenotype/Emotion_questionnaires_summary.tsv", sep="\t")
    dem = pd.read_csv(f"{bids}/phenotype/Demographics.tsv", sep="\t")[["participant_id", "gender"]]
    num = lambda c: pd.to_numeric(q[c], errors="coerce")
    q["cesd"] = pd.concat([num("cesd_Total_w1"), num("cesd_Total_w2")], axis=1).mean(axis=1)
    q["tai"] = pd.concat([num("tai_Total_w1"), num("tai_Total_w2")], axis=1).mean(axis=1)
    for s in ("depression", "fatigue", "vigour"):
        q[f"poms_{s}"] = pd.concat([num(f"poms_{s}_w1"), num(f"poms_{s}_w2")], axis=1).mean(axis=1)
    q["older"] = (q["age_group"] == "OA").astype(float)
    F = pd.read_csv(feat).merge(q[["participant_id", "cesd", "tai", "poms_depression", "poms_fatigue", "poms_vigour", "older"]], on="participant_id")
    F = F.merge(dem, on="participant_id", how="left")
    F["female"] = (F["gender"] == "F").astype(float)
    F = F[F["cesd"].notna()]
    for sig in ("hr", "sbp"):
        F[f"{sig}_reactivity"] = F[f"{sig}_stress"] - F[f"{sig}_baseline"]
        F[f"{sig}_ar1_change"] = F[f"{sig}_ar1_recovery"] - F[f"{sig}_ar1_baseline"]
    base = ["older", "female"]
    res = {"n_with_cesd": int(len(F)), "n_hr_ok": int(F["hr_ok"].sum()), "n_sbp_ok": int(F["sbp_ok"].sum()),
           "cesd_median": float(F["cesd"].median()), "cesd_ge16": int((F["cesd"] >= 16).sum())}
    P = {}
    for sig in ("hr", "sbp"):
        d = F[F[f"{sig}_ok"] == True]  # noqa: E712
        P[f"{sig}_late_recovery"] = rank_reg(d, f"{sig}_late_recovery", "cesd", [f"{sig}_baseline", f"{sig}_stress"] + base)
        P[f"{sig}_ar1_change"] = rank_reg(d, f"{sig}_ar1_change", "cesd", base)
    for k, v in holm({k: v["p_one_sided"] for k, v in P.items()}).items():
        P[k]["p_holm"] = v
    res["primary"] = P
    S = {}
    for sig in ("hr", "sbp"):
        d = F[F[f"{sig}_ok"] == True]  # noqa: E712
        S[f"{sig}_late_recovery_adj_tai"] = rank_reg(d, f"{sig}_late_recovery", "cesd", [f"{sig}_baseline", f"{sig}_stress", "tai"] + base)
        S[f"{sig}_ar1_change_adj_tai"] = rank_reg(d, f"{sig}_ar1_change", "cesd", ["tai"] + base)
        S[f"{sig}_early_recovery"] = rank_reg(d, f"{sig}_early_recovery", "cesd", [f"{sig}_baseline", f"{sig}_stress"] + base)
        S[f"{sig}_reactivity_blunting"] = rank_reg(d, f"{sig}_reactivity", "cesd", [f"{sig}_baseline"] + base, one_sided_positive=False)
        for pred in ("poms_fatigue", "poms_vigour", "poms_depression"):
            S[f"{sig}_late_recovery_{pred}"] = rank_reg(d, f"{sig}_late_recovery", pred, [f"{sig}_baseline", f"{sig}_stress"] + base,
                                                        one_sided_positive=(pred != "poms_vigour"))
            S[f"{sig}_ar1_change_{pred}"] = rank_reg(d, f"{sig}_ar1_change", pred, base, one_sided_positive=(pred != "poms_vigour"))
        hi, lo = d[d["cesd"] >= 16], d[d["cesd"] < 16]
        for y in (f"{sig}_late_recovery", f"{sig}_ar1_change"):
            a, b = hi[y].dropna(), lo[y].dropna()
            u, p = stats.mannwhitneyu(a, b)
            S[f"{y}_cesd16_auc"] = {"auc": round(float(u / (len(a) * len(b))), 4), "p_two_sided": float(p), "n_high": int(len(a)), "n_low": int(len(b))}
    # load versus mood: fatigue-minus-vigour composite versus POMS depression, bootstrap difference of standardised betas
    rng = np.random.default_rng(SEED)
    for sig in ("hr", "sbp"):
        d = F[F[f"{sig}_ok"] == True].copy()  # noqa: E712
        d["load"] = stats.zscore(d["poms_fatigue"], nan_policy="omit") - stats.zscore(d["poms_vigour"], nan_policy="omit")
        y = f"{sig}_late_recovery"; covs = [f"{sig}_baseline", f"{sig}_stress"] + base
        diffs = []
        for _ in range(NBOOT):
            b = d.sample(len(d), replace=True, random_state=int(rng.integers(1 << 31)))
            diffs.append(rank_reg(b, y, "load", covs)["beta_std"] - rank_reg(b, y, "poms_depression", covs)["beta_std"])
        obs = rank_reg(d, y, "load", covs)["beta_std"] - rank_reg(d, y, "poms_depression", covs)["beta_std"]
        S[f"{sig}_load_minus_mood_late_recovery"] = {"difference": round(obs, 4), "ci95": [round(float(np.percentile(diffs, 2.5)), 4), round(float(np.percentile(diffs, 97.5)), 4)]}
    res["secondary"] = S
    res["descriptive"] = {f"{sig}_{k}": round(float(F[F[f'{sig}_ok'] == True][f"{sig}_{k}"].median()), 3)  # noqa: E712
                          for sig in ("hr", "sbp") for k in ("baseline", "stress", "early_recovery", "late_recovery", "ar1_baseline", "ar1_recovery")}
    with open(out, "w") as fh:
        json.dump(res, fh, indent=2, sort_keys=True); fh.write("\n")


if __name__ == "__main__":
    main()
