#!/usr/bin/env python3
"""
optuna_tail_tpe_space.py — 꼬리 Optuna 연속(TPE) 탐색 공간
기존 ``_tail_grids()`` categorical 유지. ``skip_hts_scan_dupes=False`` 고정.
비율 축은 엔진과 동일(소수, UI% 아님) — evaluate/apply 경로와 맞춤.

multivariate TPE: 모든 축을 **먼저** suggest 한 뒤, 제약(ATR min/max·래칫 오름차순 등)만
맨 끝에서 TrialPruned. 중간 prune 시 뒤쪽 키가 trial마다 빠져 independent sampling 경고가 난다.
"""
from __future__ import annotations

from typing import Any, Dict, List, Optional

import optuna

from kis_trader.backtest.optuna_tpe_common import (
    RATCHET_TPE_AXIS_KEYS,
    r1,
    r2,
    r3,
    r4,
    suggest_ratchet_tiers_pct,
)

# 래칫: 꼬리는 소폭 % (기존 메뉴 0.3~2.0 대역) — 숫자축 + 조립 문자열
# 휩쏘: 실매 DB 기본(enabled=false, subbar=30, lookback=90, dip=0.003, recovery=0.001) 그리드 포함
_TAIL_BASE_KEYS: List[str] = [
    "entry_mode",
    "cand_limit",
    "max_daily_change",
    "min_drop_rate",
    "min_recovery_ratio",
    "tail_ratio_min",
    "tail_pct_min",
    "max_rec_3m",
    "shoulder_min_high",
    "shoulder_cut_pct",
    "stop_atr_mult",
    "target_atr_mult",
    "atr_sl_min_pct",
    "atr_sl_max_pct",
    "atr_tp_min_pct",
    "atr_tp_max_pct",
    "tail_vol_mult",
    "tail_vol_win",
    "limit_atr_mult",
    "symbol_daily_loss_limit_pct",
    "symbol_daily_loss_limit_krw",
    "reentry_min_edge_krw",
    "reentry_require_nonneg",
    "max_daily",
    "cooldown_min",
    "bar_chg_min_pct",
    "bar_chg_max_pct",
    "rsi_threshold",
    "pattern_pin",
    "pattern_engulfing",
    "pattern_piercing",
    "pattern_harami",
    "pattern_doji",
    "pattern_morning_star",
    # 당일손익 다단트레일(trail_tiers/drop/arm) — 운영 리스크 손잡이. TPE·apply 탐색 제외(DB/UI 고정).
    "max_loss_krw",
    "whipsaw_enabled",
    "whipsaw_subbar_sec",
    "whipsaw_lookback_sec",
    "whipsaw_dip_pct",
    "whipsaw_recovery_tol_pct",
]
TAIL_TPE_AXIS_KEYS: List[str] = list(_TAIL_BASE_KEYS) + list(RATCHET_TPE_AXIS_KEYS)


def tail_tpe_axis_keys() -> List[str]:
    return list(TAIL_TPE_AXIS_KEYS)


def normalize_tpe_tail_entry_mode(raw: Optional[Any] = None) -> str:
    """TPE는 진입모드를 탐색하지 않고 스터디마다 고정. align | limit_atr."""
    s = str(raw or "").strip().lower()
    if s in ("limit_atr", "limit", "atr_limit"):
        return "limit_atr"
    return "align"


def suggest_tail_params_tpe(
    trial: optuna.Trial,
    entry_mode: Optional[str] = None,
) -> Dict[str, Any]:
    combo: Dict[str, Any] = {}
    # 한 스터디=한 모드. categorical 혼입 금지(웹 체크 2개면 잡 2개 순차).
    combo["entry_mode"] = normalize_tpe_tail_entry_mode(entry_mode)
    combo["cand_limit"] = trial.suggest_categorical("cand_limit", [0, 20])
    combo["max_daily_change"] = r1(
        trial.suggest_float("max_daily_change", 10.0, 60.0, step=1.0),  # 확장: 20→10, 45→60
    )
    combo["min_drop_rate"] = r3(
        trial.suggest_float("min_drop_rate", 0.005, 0.15, step=0.005),  # 확장: 0.01→0.005, 0.08→0.15
    )
    combo["min_recovery_ratio"] = r2(
        trial.suggest_float("min_recovery_ratio", 0.08, 0.35, step=0.01),
    )
    combo["tail_ratio_min"] = r2(trial.suggest_float("tail_ratio_min", 0.4, 1.5, step=0.1))
    combo["tail_pct_min"] = r4(
        trial.suggest_float("tail_pct_min", 0.0005, 0.01, step=0.0005),
    )
    combo["max_rec_3m"] = r2(trial.suggest_float("max_rec_3m", 0.7, 0.98, step=0.01))
    combo["shoulder_min_high"] = r4(
        trial.suggest_float("shoulder_min_high", 0.002, 0.015, step=0.001),
    )
    combo["shoulder_cut_pct"] = r4(
        trial.suggest_float("shoulder_cut_pct", 0.0005, 0.005, step=0.0005),
    )
    combo["stop_atr_mult"] = r2(trial.suggest_float("stop_atr_mult", 0.8, 3.0, step=0.1))
    combo["target_atr_mult"] = r2(
        trial.suggest_float("target_atr_mult", 0.8, 3.5, step=0.1),
    )
    combo["atr_sl_min_pct"] = r2(trial.suggest_float("atr_sl_min_pct", 0.3, 1.0, step=0.1))
    combo["atr_sl_max_pct"] = r1(trial.suggest_float("atr_sl_max_pct", 1.0, 12.0, step=0.5))  # 확장: 8.0→12.0
    combo["atr_tp_min_pct"] = r2(trial.suggest_float("atr_tp_min_pct", 0.1, 2.0, step=0.1))  # 확장: 0.2→0.1
    combo["atr_tp_max_pct"] = r1(trial.suggest_float("atr_tp_max_pct", 1.0, 10.0, step=0.5))  # 확장: 6.0→10.0

    combo["tail_vol_mult"] = r2(trial.suggest_float("tail_vol_mult", 0.0, 4.0, step=0.1))
    combo["tail_vol_win"] = trial.suggest_int("tail_vol_win", 2, 8)
    combo["limit_atr_mult"] = r2(trial.suggest_float("limit_atr_mult", 0.8, 2.5, step=0.1))
    # 꼬리 래칫: OFF=\"off\" / gain·cut 소폭% (엔진 문자열과 동일)
    combo.update(
        suggest_ratchet_tiers_pct(
            trial,
            off_token="off",
            n_max=3,
            gain_low=0.3,
            gain_high=3.0,
            gain_step=0.1,
            cut_low=0.15,
            cut_high=0.5,
            cut_step=0.05,
        ),
    )
    combo["symbol_daily_loss_limit_pct"] = r2(
        trial.suggest_float("symbol_daily_loss_limit_pct", 0.0, 3.0, step=0.5),
    )
    combo["symbol_daily_loss_limit_krw"] = trial.suggest_int(
        "symbol_daily_loss_limit_krw", 0, 80000, step=10000,
    )
    combo["reentry_min_edge_krw"] = trial.suggest_int(
        "reentry_min_edge_krw", 0, 1000, step=100,
    )
    combo["reentry_require_nonneg"] = False
    combo["max_daily"] = trial.suggest_int("max_daily", 3, 80, step=5)  # 확장: 5→3, 60→80
    combo["cooldown_min"] = r1(trial.suggest_float("cooldown_min", 0.0, 30.0, step=1.0))  # 확장: 15→30
    combo["bar_chg_min_pct"] = r1(
        trial.suggest_float("bar_chg_min_pct", -15.0, -3.0, step=0.5),
    )
    combo["bar_chg_max_pct"] = r2(
        trial.suggest_float("bar_chg_max_pct", -2.0, -0.2, step=0.1),
    )

    combo["rsi_threshold"] = r1(trial.suggest_float("rsi_threshold", 70.0, 95.0, step=1.0))
    for pk in (
        "pattern_pin",
        "pattern_engulfing",
        "pattern_piercing",
        "pattern_harami",
        "pattern_doji",
        "pattern_morning_star",
    ):
        combo[pk] = trial.suggest_categorical(pk, [False, True])

    combo["max_loss_krw"] = trial.suggest_int("max_loss_krw", 50000, 300000, step=25000)

    # 휩쏘 TRIGGER — 실매값(False/30/90/0.003/0.001) 포함. ON·OFF·초·% 축 분리.
    combo["whipsaw_enabled"] = trial.suggest_categorical(
        "whipsaw_enabled", [False, True],
    )
    combo["whipsaw_subbar_sec"] = trial.suggest_categorical(
        "whipsaw_subbar_sec", [15, 30, 45, 60],
    )
    combo["whipsaw_lookback_sec"] = trial.suggest_categorical(
        "whipsaw_lookback_sec", [60, 90, 120, 180],
    )
    combo["whipsaw_dip_pct"] = r4(
        trial.suggest_float("whipsaw_dip_pct", 0.001, 0.01, step=0.001),
    )
    combo["whipsaw_recovery_tol_pct"] = r4(
        trial.suggest_float("whipsaw_recovery_tol_pct", 0.0005, 0.003, step=0.0005),
    )

    combo["skip_hts_scan_dupes"] = False

    # --- 제약 prune: 모든 suggest 이후에만 (키 공간 고정) ---
    if combo["atr_sl_min_pct"] >= combo["atr_sl_max_pct"]:
        raise optuna.TrialPruned("atr_sl min>=max")
    if combo["atr_tp_min_pct"] >= combo["atr_tp_max_pct"]:
        raise optuna.TrialPruned("atr_tp min>=max")
    if combo["bar_chg_min_pct"] >= combo["bar_chg_max_pct"]:
        raise optuna.TrialPruned("bar_chg invalid")
    if combo.pop("_ratchet_ascending_ok", True) is False:
        raise optuna.TrialPruned("ratchet gain not ascending")

    return combo