"""Stochastic depth (DropPath) fine-tune of stem_ms96 from SWA checkpoints.

Huang et al. / DeiT-style DropPath on transformer residual branches of the
published abs-pos stem_ms96 recipe (dilated stem + 24-d ms + w96). Distinct
from v4 `rel_ms` relative-bias local attention which already used stochdepth
0.1 — this patches FFN (and optionally attention) residuals on the SWA abs-pos
encoder while keeping absolute `pos`.

Init: models/stem_ms96_swa/{held-*,all5}.pt (IRM only if env INIT=irm).
Eval: EMA decay 0.995. HMM preset p_stay 0.994 scale 0.8. Also report causal
tau-30 high-pass of the evidence. Never uses AASD/Carta labels.

usage:
  DROP_PATH=0.1 STEPS=1000 python stem_ms96_stochdepth.py
  TUNE_DROP=1 STEPS=1000 python stem_ms96_stochdepth.py   # smoke p then full
  SMOKE=1 STEPS=40 python stem_ms96_stochdepth.py         # smoke only
"""
import copy
import json
import math
import os
import pickle
import time
import types

import numpy as np
import torch
import torch.nn.functional as F

import ensemble_clf_nested as N
import ensemble_clf_transformer_v2 as V
import ensemble_clf_transformer_v4 as V4
import eval_external as E
from common import DS_KEYS

EXP = os.path.dirname(os.path.abspath(__file__))
INIT_NAME = os.environ.get("INIT", "swa").strip().lower()  # swa | irm
INIT_DIR = os.path.join(EXP, "models", f"stem_ms96_{INIT_NAME}")
CKPT_DIR = os.path.join(EXP, "models", "stem_ms96_stochdepth")
LOG = os.path.join(EXP, "experiments", "stem_ms96_stochdepth.log")
OUT = os.path.join(EXP, "results", "research", "stem_ms96_stochdepth.json")
EVID = "/root/disk_3tb/EEG-sAAD-Grand-Challenge-Training-Dataset/experiments/decoder_lodo/results/research"
TEST_NAMES = ("AASD", "Carta2026")

STEPS = int(os.environ.get("STEPS", "1000"))
BATCH = int(os.environ.get("BATCH", "512"))
DROP_PATH = float(os.environ.get("DROP_PATH", "0.1"))
ATTN_DROP = os.environ.get("ATTN_DROP", "1") == "1"  # DropPath on attn residual too
SMOKE = os.environ.get("SMOKE", "0") == "1"
TUNE_DROP = os.environ.get("TUNE_DROP", "0") == "1"
EMA_DECAY = 0.995
LR = 8e-4
P_STAY = 0.994
SCALE = 0.8
CFG = ("stem_ms96", "ms", 96, 64, 2, True, LR)
WINDOW = 96
CHANCE_LOSS = 0.685


def log_line(text):
    print(text, flush=True)
    os.makedirs(os.path.dirname(LOG), exist_ok=True)
    with open(LOG, "a") as fh:
        fh.write(text + "\n")


def drop_path(x, drop_prob, training):
    """Per-sample residual DropPath (keep-prob scaling)."""
    if drop_prob <= 0.0 or not training:
        return x
    keep = 1.0 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)
    mask = x.new_empty(shape).bernoulli_(keep)
    return x.mul(mask).div_(keep)


def _patch_layer_forward(layer, drop_attn, drop_ffn):
    """Replace TransformerEncoderLayer.forward with DropPath residuals.

    norm_first Pre-LN: x = x + DP(sa(norm1(x))); x = x + DP(ff(norm2(x))).
    """

    def forward(self, src, src_mask=None, src_key_padding_mask=None, is_causal=False):
        x = src
        sa = self._sa_block(self.norm1(x), src_mask, src_key_padding_mask, is_causal=is_causal)
        x = x + drop_path(sa, drop_attn, self.training)
        ff = self._ff_block(self.norm2(x))
        x = x + drop_path(ff, drop_ffn, self.training)
        return x

    layer.forward = types.MethodType(forward, layer)


def apply_stochdepth(model, drop_path, attn_drop=True):
    """Linear depth schedule: layer i uses p * (i+1)/N on FFN (and attn if set)."""
    layers = list(model.enc.layers)
    n = max(len(layers), 1)
    rates = []
    for i, layer in enumerate(layers):
        p = float(drop_path) * float(i + 1) / float(n)
        drop_attn = p if attn_drop else 0.0
        drop_ffn = p
        _patch_layer_forward(layer, drop_attn, drop_ffn)
        rates.append(dict(layer=i, drop_attn=drop_attn, drop_ffn=drop_ffn))
    model._stochdepth_rates = rates
    model._stochdepth_p = float(drop_path)
    model._stochdepth_attn = bool(attn_drop)
    return rates


def load_init(path, device, drop_path, attn_drop):
    blob = torch.load(path, map_location="cpu", weights_only=False)
    name, family, window, d_model, nlayers, stem, _lr = blob["cfg"]
    if name != "stem_ms96" or not stem or window != WINDOW:
        raise RuntimeError(f"unexpected stem cfg {blob['cfg']}")
    c_in = int(np.asarray(blob["mu"]).shape[0])
    net = V.LogitTransformer(c_in, window, d_model=d_model, nlayers=nlayers, stem=True)
    good_mask = net.mask.detach().clone()
    net.load_state_dict(blob["state"])
    net.mask.copy_(good_mask)
    if torch.isnan(net.mask).any():
        raise RuntimeError(f"causal mask still has nan after restore from {path}")
    rates = apply_stochdepth(net, drop_path, attn_drop=attn_drop)
    net = net.to(device)
    mu = np.asarray(blob["mu"], np.float32)
    sd = np.asarray(blob["sd"], np.float32)
    return net, mu, sd, list(blob["train"]), rates


def weighted_bce(logit, y, m, pos_weight):
    sw = V.switch_weights(y, m)
    bce = F.binary_cross_entropy_with_logits(logit, y, pos_weight=pos_weight, reduction="none")
    denom = (sw * m).sum().clamp_min(1.0)
    loss = (bce * sw * m).sum() / denom
    dv = (logit[:, 1:] - logit[:, :-1]).abs()
    m2 = m[:, 1:] * m[:, :-1]
    loss = loss + 0.05 * (dv * m2).sum() / m2.sum().clamp_min(1.0)
    return loss


def set_lr(opt, step, total, base):
    warm = min(40, max(total // 10, 1))
    if step < warm:
        lr = base * (step + 1) / warm
    else:
        p = (step - warm) / max(total - warm, 1)
        lr = 1e-5 + 0.5 * (base - 1e-5) * (1.0 + math.cos(math.pi * p))
    for group in opt.param_groups:
        group["lr"] = lr
    return lr


def train_stoch(model, groups, y_all, mu, sd, device, steps, batch, tag, abort_after=50):
    pos = float(y_all.mean())
    pos_weight = torch.tensor([(1.0 - pos) / max(pos, 1e-3)], device=device)
    opt = torch.optim.AdamW(model.parameters(), lr=LR, weight_decay=1e-4)
    rng = np.random.default_rng(V.TRAIN_SEED)
    ema = copy.deepcopy(model).eval()
    # EMA must keep DropPath patched (eval → drop_path no-op) for identical forward
    apply_stochdepth(ema, getattr(model, "_stochdepth_p", 0.0),
                     attn_drop=getattr(model, "_stochdepth_attn", True))
    ema.eval()
    model.train()
    t0 = time.time()
    last = None
    early_losses = []
    p = float(getattr(model, "_stochdepth_p", 0.0))
    for step in range(steps):
        set_lr(opt, step, steps, LR)
        xb, yb, mb = V.sample_batch(rng, groups, WINDOW, batch, mu, sd)
        x = torch.from_numpy(xb).to(device, non_blocking=True)
        y = torch.from_numpy(yb).to(device, non_blocking=True)
        m = torch.from_numpy(mb).to(device, non_blocking=True)
        x = V.augment(x)
        opt.zero_grad(set_to_none=True)
        loss = weighted_bce(model(x), y, m, pos_weight)
        if not torch.isfinite(loss):
            raise RuntimeError(f"non-finite loss at step {step} for {tag}")
        loss.backward()
        torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)
        opt.step()
        V.ema_update(model, ema, EMA_DECAY)
        last = float(loss.item())
        if step < abort_after:
            early_losses.append(last)
            if step + 1 == abort_after:
                mean_early = float(np.mean(early_losses))
                log_line(f"    {tag} early-abort check mean_loss@{abort_after}={mean_early:.4f}")
                if mean_early >= CHANCE_LOSS:
                    raise RuntimeError(
                        f"early abort {tag}: mean loss {mean_early:.4f} >= chance {CHANCE_LOSS}")
        if step == 0 or (step + 1) % 200 == 0 or step + 1 == steps:
            log_line(f"    {tag} step {step + 1}/{steps} loss {last:.4f} drop_path {p}")
    ema.eval()
    acc = V.train_frame_acc(ema, groups, WINDOW, mu, sd, device)
    log_line(
        f"  trained {tag} loss {last:.4f} frame-acc {acc:.3f} drop_path {p} "
        f"({time.time() - t0:.0f}s)")
    return ema, dict(loss=last, frame_acc=acc, drop_path=p,
                     attn_drop=bool(getattr(model, "_stochdepth_attn", True)),
                     rates=getattr(model, "_stochdepth_rates", None))


def fit_with_oom(init_path, groups, y_all, device, steps, batch, tag, drop_path,
                 attn_drop, halved=False):
    model, mu, sd, _train, rates = load_init(init_path, device, drop_path, attn_drop)
    log_line(f"  {tag} stochdepth rates {rates}")
    try:
        ema, stats = train_stoch(model, groups, y_all, mu, sd, device, steps, batch, tag)
        return ema, mu, sd, batch, stats
    except torch.cuda.OutOfMemoryError:
        if halved:
            raise
        log_line(f"  OOM on {tag} at batch {batch}; halving once")
        del model
        if device.type == "cuda":
            torch.cuda.empty_cache()
        return fit_with_oom(
            init_path, groups, y_all, device, steps, batch // 2, tag, drop_path,
            attn_drop, halved=True)


def hp30_stream(stream):
    out = {}
    for key, d in stream.items():
        out[key] = V4.highpass(np.asarray(d, np.float64)[:, None]).reshape(-1).astype(np.float64)
    return out


def score_lodo(model, feat, mu, sd, held, device):
    streams = V.infer_all(model, feat, mu, sd, WINDOW, device)
    ev = N.to_ev(streams)
    off = N.official(held, ev, P_STAY, SCALE)
    hp = {ds: hp30_stream(streams[ds]) for ds in streams}
    ev_hp = N.to_ev(hp)
    off_hp = N.official(held, ev_hp, P_STAY, SCALE)
    log_line(f"LODO {held} p {P_STAY} sc {SCALE}  H {off['H']:.4f} Acc {off['Acc']:.2f} "
             f"L {off['L']:.3f} SDT {off['SDT']:.2f}")
    log_line(f"LODO_hp30 {held} p {P_STAY} sc {SCALE}  H {off_hp['H']:.4f} Acc {off_hp['Acc']:.2f}")
    return dict(stoch=off, hp30=off_hp), streams


def score_external(model, feat_ext, meta, mu_in, sd_in, corpus_streams, corpus_att, device):
    parts = []
    for ds, table in corpus_streams.items():
        labels = corpus_att[ds]
        for key, d in table.items():
            parts.append(np.asarray(d, np.float64) * np.where(labels[key] == 0, 1.0, -1.0))
    mu_e = float(np.concatenate(parts).mean())
    sd_e = float(np.concatenate(parts).std())
    log_line(f"  external emissions mu {mu_e:.4f} sd {sd_e:.4f} scale {SCALE}")

    hp_corpus = {ds: hp30_stream(corpus_streams[ds]) for ds in corpus_streams}
    parts_hp = []
    for ds, table in hp_corpus.items():
        labels = corpus_att[ds]
        for key, d in table.items():
            parts_hp.append(np.asarray(d, np.float64) * np.where(labels[key] == 0, 1.0, -1.0))
    mu_hp = float(np.concatenate(parts_hp).mean())
    sd_hp = float(np.concatenate(parts_hp).std())
    log_line(f"  external hp30 emissions mu {mu_hp:.4f} sd {sd_hp:.4f} scale {SCALE}")

    rows = {}
    for ds in TEST_NAMES:
        stream = V.infer_all(model, {ds: feat_ext[ds]}, mu_in, sd_in, WINDOW, device)[ds]
        ev = {}
        for key, d in stream.items():
            meta_k = meta[ds][key]
            d = np.asarray(d, np.float64)
            ev[key] = dict(r0=d, r1=np.zeros(len(d)), att=meta_k["att"], t_end=meta_k["t_end"],
                           att128=meta_k["att128"], n128=meta_k["n128"], subject=meta_k["subject"])
        r = E.score(ev, mu_e, sd_e * SCALE, P_STAY, 0.75)
        stream_hp = hp30_stream(stream)
        ev_hp = {}
        for key, d in stream_hp.items():
            meta_k = meta[ds][key]
            d = np.asarray(d, np.float64)
            ev_hp[key] = dict(r0=d, r1=np.zeros(len(d)), att=meta_k["att"], t_end=meta_k["t_end"],
                              att128=meta_k["att128"], n128=meta_k["n128"], subject=meta_k["subject"])
        r_hp = E.score(ev_hp, mu_hp, sd_hp * SCALE, P_STAY, 0.75)
        rows[ds] = dict(stoch=r, hp30=r_hp)
        log_line(f"EXT {ds} p {P_STAY} sc {SCALE}  H {r['H']:.4f} Acc {r['Acc']:.2f} "
                 f"L {r['L']:.3f} SDT {r['SDT']:.2f}")
        log_line(f"EXT_hp30 {ds} p {P_STAY} sc {SCALE}  H {r_hp['H']:.4f} Acc {r_hp['Acc']:.2f}")
    return rows


def load_external():
    E.CONF.update(prep="eaP_n5_t60", targets=["env", "onset", "dom"])
    logits, meta = {}, {}
    for ds in TEST_NAMES:
        evs = []
        for i in range(4):
            path = os.path.join(EVID, f"ext_clf_{ds}_w10b3sw2P_evidence_s{i}.pkl")
            with open(path, "rb") as fh:
                evs.append(pickle.load(fh))
        logits[ds], meta[ds] = {}, {}
        for key in evs[0]:
            cols = [(ev[key]["r0"] - ev[key]["r1"]).astype(np.float32) for ev in evs]
            e0 = evs[0][key]
            logits[ds][key] = np.stack(cols, 1)
            meta[ds][key] = dict(
                att=np.asarray(e0["att"]), t_end=np.asarray(e0["t_end"], np.float64),
                att128=np.asarray(e0["att128"]), n128=int(e0["n128"]), subject=int(e0["subject"]))
        log_line(f"  external {ds}: {len(logits[ds])} trials")
    return logits, meta


def smoke_drop(feat_all, device, attn_drop):
    """Short train on held-AVGC for p∈{0.05,0.1,0.2}; pick finite + highest frame_acc."""
    held = "AVGC"
    path = os.path.join(INIT_DIR, f"held-{held}.pt")
    train_ds = [d for d in DS_KEYS if d != held]
    groups, y_all = V.trial_groups(feat_all, train_ds)
    candidates = [float(x) for x in os.environ.get("SMOKE_DROPS", "0.05,0.1,0.2").split(",")]
    smoke_steps = int(os.environ.get("SMOKE_STEPS", "40"))
    results = []
    for p in candidates:
        log_line(f"SMOKE drop_path={p} attn_drop={attn_drop} steps={smoke_steps}")
        try:
            ema, mu, sd, used, stats = fit_with_oom(
                path, groups, y_all, device, smoke_steps, BATCH, f"smoke-p{p}", p, attn_drop)
            del ema
            if device.type == "cuda":
                torch.cuda.empty_cache()
            ok = (stats["loss"] is not None and math.isfinite(stats["loss"])
                  and stats["frame_acc"] > 0.55 and stats["loss"] < CHANCE_LOSS)
            results.append(dict(ok=ok, batch=used, **stats))
            log_line(f"SMOKE drop_path={p} ok={ok} loss={stats['loss']:.4f} "
                     f"frame={stats['frame_acc']:.3f}")
        except Exception as exc:
            log_line(f"SMOKE drop_path={p} FAILED: {exc}")
            results.append(dict(drop_path=p, ok=False, error=str(exc)))
            if device.type == "cuda":
                torch.cuda.empty_cache()
    ok_rows = [r for r in results if r.get("ok")]
    if not ok_rows:
        raise RuntimeError(f"all smoke drop_path failed: {results}")
    best = max(ok_rows, key=lambda r: (r["frame_acc"], -r["loss"]))
    log_line(f"SMOKE chose drop_path={best['drop_path']} frame={best['frame_acc']:.3f} "
             f"loss={best['loss']:.4f}")
    return float(best["drop_path"]), results


def main():
    os.makedirs(os.path.dirname(LOG), exist_ok=True)
    open(LOG, "w").close()
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    if INIT_NAME not in ("swa", "irm"):
        raise RuntimeError(f"INIT must be swa|irm, got {INIT_NAME}")
    if not os.path.isdir(INIT_DIR):
        raise FileNotFoundError(INIT_DIR)
    log_line(f"device {device} steps {STEPS} batch {BATCH} drop_path {DROP_PATH} "
             f"attn_drop={ATTN_DROP} init={INIT_NAME} smoke={SMOKE} tune_drop={TUNE_DROP}")
    torch.backends.cuda.matmul.allow_tf32 = True
    torch.backends.cudnn.allow_tf32 = True
    V.assert_causal(device)
    N.load_all()
    feat_all = V.build_features()["ms"]
    os.makedirs(CKPT_DIR, exist_ok=True)

    drop_p = DROP_PATH
    smoke_info = None
    if SMOKE or TUNE_DROP:
        drop_p, smoke_info = smoke_drop(feat_all, device, ATTN_DROP)
        if SMOKE:
            out = dict(smoke=smoke_info, chosen_drop_path=drop_p, init=INIT_NAME,
                       attn_drop=ATTN_DROP)
            os.makedirs(os.path.dirname(OUT), exist_ok=True)
            with open(OUT.replace(".json", "_smoke.json"), "w") as fh:
                json.dump(out, fh, indent=1, default=float)
            log_line("SMOKE done; exiting without full score")
            return

    per = {}
    for held in DS_KEYS:
        path = os.path.join(INIT_DIR, f"held-{held}.pt")
        if not os.path.isfile(path):
            raise FileNotFoundError(path)
        train_ds = [d for d in DS_KEYS if d != held]
        groups, y_all = V.trial_groups(feat_all, train_ds)
        log_line(f"fold held-out={held} init {path} drop_path={drop_p}")
        ema, mu, sd, used, stats = fit_with_oom(
            path, groups, y_all, device, STEPS, BATCH, f"held-{held}", drop_p, ATTN_DROP)
        torch.save({"state": ema.state_dict(), "mu": mu, "sd": sd, "init": path,
                    "train": train_ds, "steps": STEPS, "batch": used,
                    "drop_path": drop_p, "attn_drop": ATTN_DROP, "stats": stats, "cfg": CFG},
                   os.path.join(CKPT_DIR, f"held-{held}.pt"))
        scores, _streams = score_lodo(ema, feat_all, mu, sd, held, device)
        per[held] = dict(scores=scores, stats=stats, batch=used)
        del ema
        if device.type == "cuda":
            torch.cuda.empty_cache()

    log_line(f"all-5 stochdepth fine-tune for AASD and Carta, init {INIT_DIR}/all5.pt "
             f"drop_path={drop_p}")
    groups, y_all = V.trial_groups(feat_all, list(DS_KEYS))
    all_path = os.path.join(INIT_DIR, "all5.pt")
    ema, mu, sd, used, stats = fit_with_oom(
        all_path, groups, y_all, device, STEPS, BATCH, "all5", drop_p, ATTN_DROP)
    torch.save({"state": ema.state_dict(), "mu": mu, "sd": sd, "init": all_path,
                "train": list(DS_KEYS), "steps": STEPS, "batch": used,
                "drop_path": drop_p, "attn_drop": ATTN_DROP, "stats": stats, "cfg": CFG},
               os.path.join(CKPT_DIR, "all5.pt"))
    corpus_streams = V.infer_all(ema, feat_all, mu, sd, WINDOW, device)
    ext_logits, ext_meta = load_external()
    ext_feat = V4.build_features_from(ext_logits)["ms"]
    ext = score_external(ema, ext_feat, ext_meta, mu, sd, corpus_streams, N.G["att"], device)

    mean_h = float(np.mean([per[d]["scores"]["stoch"]["H"] for d in DS_KEYS]))
    mean_acc = float(np.mean([per[d]["scores"]["stoch"]["Acc"] for d in DS_KEYS]))
    log_line(f"LODO mean H {mean_h:.4f} Acc {mean_acc:.2f}")

    out = dict(
        p_stay=P_STAY, scale=SCALE, steps=STEPS, batch=BATCH,
        drop_path=drop_p, attn_drop=ATTN_DROP, init=INIT_NAME, smoke=smoke_info,
        lodo={d: per[d]["scores"] for d in DS_KEYS},
        lodo_stats={d: per[d]["stats"] for d in DS_KEYS},
        lodo_mean=dict(H=mean_h, Acc=mean_acc),
        external=ext,
        all5_stats=stats,
    )
    os.makedirs(os.path.dirname(OUT), exist_ok=True)
    with open(OUT, "w") as fh:
        json.dump(out, fh, indent=1, default=float)
    log_line(f"wrote {OUT}")


if __name__ == "__main__":
    main()
