"""Denser synthetic-switch fine-tune of stem_ms96 from SWA (abs-pos kept).

Same LogitTransformer as published stem_ms96 / SWA (absolute pos, dilated stem,
ms features, window 96). Training difference only:

  * With P_SWITCH, insert 1 synthetic attention switch inside the w96 window
    (flip labels after tau; negate all ms channels after tau ≈ stream swap).
  * With P_SWITCH2 among switched rows, insert a second switch-back ≥ MIN_GAP
    frames later (multi-switch windows).
  * Stronger post-switch BCE weights (SW_GAIN vs recipe 2.5).

NOT RoPE / ALiBi / CORAL / fuse↔hp30. Distinct from classifier technique-synth-
switch (10 s vlaai + mix820 trial stamps).

Init models/stem_ms96_swa/{held-*,all5}.pt; restore causal -inf mask. EMA 0.995.
Score LODO 5 + AASD + Carta at HMM p_stay=0.994 scale=0.8; also hp30.

usage:
  STEPS=1000 BATCH=512 python stem_ms96_synth_switch.py
"""
import copy
import json
import math
import os
import pickle
import time

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__))
SWA_DIR = os.path.join(EXP, "models", "stem_ms96_swa")
STEM_DIR = os.path.join(EXP, "models", "logit_tf_v2")
CKPT_DIR = os.path.join(EXP, "models", "stem_ms96_synth_switch")
LOG = os.path.join(EXP, "experiments", "stem_ms96_synth_switch.log")
OUT = os.path.join(EXP, "results", "research", "stem_ms96_synth_switch.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"))
EMA_DECAY = 0.995
LR = 8e-4
P_STAY = 0.994
SCALE = 0.8
# denser than classifier sw2 defaults (0.5 / 0.5) and recipe post-switch gain 2.5
P_SWITCH = float(os.environ.get("P_SWITCH", "0.7"))
P_SWITCH2 = float(os.environ.get("P_SWITCH2", "0.7"))
MIN_GAP = int(os.environ.get("MIN_GAP", "8"))  # frames (~2 s at 4 Hz)
SW_GAIN = float(os.environ.get("SW_GAIN", "5.0"))
SW_HORIZON = int(os.environ.get("SW_HORIZON", "12"))
SW_TAU = float(os.environ.get("SW_TAU", "4.0"))
CFG = ("stem_ms96", "ms", 96, 64, 2, True, LR)
WINDOW = 96


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 resolve_init(held):
    if held == "all5":
        swa = os.path.join(SWA_DIR, "all5.pt")
        if os.path.isfile(swa):
            return swa
        stem = os.path.join(STEM_DIR, "stem_ms96__held-AVGC.pt")
        if os.path.isfile(stem):
            log_line(f"  SWA all5 missing; falling back to {stem}")
            return stem
        raise FileNotFoundError("no all5 init")
    swa = os.path.join(SWA_DIR, f"held-{held}.pt")
    if os.path.isfile(swa):
        return swa
    stem = os.path.join(STEM_DIR, f"stem_ms96__held-{held}.pt")
    if os.path.isfile(stem):
        log_line(f"  SWA missing for {held}; falling back to {stem}")
        return stem
    raise FileNotFoundError(f"no init for {held}")


def load_swa(path, device):
    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 int(window) != WINDOW:
        raise RuntimeError(f"unexpected stem cfg {blob['cfg']} from {path}")
    c_in = int(np.asarray(blob["mu"]).shape[0])
    net = V.LogitTransformer(c_in, window, d_model=d_model, nlayers=nlayers, stem=True)
    # SWA buffer avg can turn causal -inf into nan; restore fresh mask.
    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}")
    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"])


def switch_weights_strong(y, mask, gain=SW_GAIN, horizon=SW_HORIZON, tau=SW_TAU):
    """Like V.switch_weights but with configurable post-switch gain (recipe uses 2.5)."""
    change = torch.zeros_like(y)
    change[:, 1:] = (y[:, 1:] - y[:, :-1]).abs().gt(0.5).float() * mask[:, 1:] * mask[:, :-1]
    decay = torch.exp(-torch.arange(horizon, device=y.device, dtype=y.dtype) / tau)
    kernel = decay.flip(0).view(1, 1, -1)
    smooth = F.conv1d(F.pad(change.unsqueeze(1), (horizon - 1, 0)), kernel).squeeze(1)
    return 1.0 + gain * smooth


def apply_synth_switches(rng, x, y, m, p_switch=P_SWITCH, p_switch2=P_SWITCH2,
                         min_gap=MIN_GAP):
    """In-place denser synthetic switches on a numpy batch (B, T, C) / (B, T).

    One switch: flip y and negate x after tau (approx. stream swap on ms feats).
    Second switch: flip back after tau2 ≥ tau + min_gap (multi-switch window).
    Only frames with mask>0 participate; early padding left alone.
    """
    b, t, _c = x.shape
    n1 = n2 = 0
    for i in range(b):
        valid = np.flatnonzero(m[i] > 0.5)
        if len(valid) < min_gap + 4:
            continue
        if rng.random() >= p_switch:
            continue
        # first switch: leave ≥2 valid frames before and after
        lo = int(valid[1])
        hi = int(valid[-2])
        if hi <= lo:
            continue
        tau = int(rng.integers(lo, hi + 1))
        x[i, tau:] *= -1.0
        y[i, tau:] = 1.0 - y[i, tau:]
        n1 += 1
        # second switch-back ≥ min_gap later, still inside the window
        if rng.random() >= p_switch2:
            continue
        tau2_lo = tau + min_gap
        if tau2_lo > hi:
            continue
        tau2 = int(rng.integers(tau2_lo, hi + 1))
        x[i, tau2:] *= -1.0
        y[i, tau2:] = 1.0 - y[i, tau2:]
        n2 += 1
    return n1, n2


def weighted_bce(logit, y, m, pos_weight):
    sw = switch_weights_strong(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_synth(model, groups, y_all, mu, sd, device, steps, batch, tag):
    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()
    model.train()
    t0 = time.time()
    last = None
    tot_sw1 = tot_sw2 = 0
    for step in range(steps):
        set_lr(opt, step, steps, LR)
        xb, yb, mb = V.sample_batch(rng, groups, WINDOW, batch, mu, sd)
        n1, n2 = apply_synth_switches(rng, xb, yb, mb)
        tot_sw1 += n1
        tot_sw2 += n2
        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)
        logit = model(x)
        loss = weighted_bce(logit, 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 == 0 or (step + 1) % 200 == 0 or step + 1 == steps:
            rate1 = tot_sw1 / max((step + 1) * batch, 1)
            rate2 = tot_sw2 / max((step + 1) * batch, 1)
            log_line(
                f"    {tag} step {step + 1}/{steps} loss {last:.4f} "
                f"sw1_rate {rate1:.3f} sw2_rate {rate2:.3f}")
    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} "
        f"sw1 {tot_sw1} sw2 {tot_sw2} ({time.time() - t0:.0f}s)")
    return ema, dict(
        loss=last, frame_acc=acc, sw1=int(tot_sw1), sw2=int(tot_sw2),
        p_switch=P_SWITCH, p_switch2=P_SWITCH2, sw_gain=SW_GAIN)


def fit_with_oom(init_path, groups, y_all, device, steps, batch, tag, halved=False):
    model, mu, sd, _train = load_swa(init_path, device)
    try:
        ema, stats = train_synth(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, 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(synth=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(synth=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 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")
    log_line(
        f"device {device} steps {STEPS} batch {BATCH} "
        f"p_switch {P_SWITCH} p_switch2 {P_SWITCH2} min_gap {MIN_GAP} "
        f"sw_gain {SW_GAIN} init {SWA_DIR}")
    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)

    per = {}
    for held in DS_KEYS:
        path = resolve_init(held)
        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}")
        ema, mu, sd, used, stats = fit_with_oom(
            path, groups, y_all, device, STEPS, BATCH, f"held-{held}")
        torch.save({"state": ema.state_dict(), "mu": mu, "sd": sd, "init": path,
                    "train": train_ds, "steps": STEPS, "batch": used,
                    "stats": stats, "cfg": CFG,
                    "p_switch": P_SWITCH, "p_switch2": P_SWITCH2,
                    "min_gap": MIN_GAP, "sw_gain": SW_GAIN},
                   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()

    all_path = resolve_init("all5")
    log_line(f"all-5 synth-switch fine-tune for AASD and Carta, init {all_path}")
    groups, y_all = V.trial_groups(feat_all, list(DS_KEYS))
    ema, mu, sd, used, stats = fit_with_oom(
        all_path, groups, y_all, device, STEPS, BATCH, "all5")
    torch.save({"state": ema.state_dict(), "mu": mu, "sd": sd, "init": all_path,
                "train": list(DS_KEYS), "steps": STEPS, "batch": used,
                "stats": stats, "cfg": CFG,
                "p_switch": P_SWITCH, "p_switch2": P_SWITCH2,
                "min_gap": MIN_GAP, "sw_gain": SW_GAIN},
               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"]["synth"]["H"] for d in DS_KEYS]))
    mean_acc = float(np.mean([per[d]["scores"]["synth"]["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,
        p_switch=P_SWITCH, p_switch2=P_SWITCH2, min_gap=MIN_GAP, sw_gain=SW_GAIN,
        init_dir=SWA_DIR,
        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()
