kis_bot/kis_trader/backtest/optuna_breakout.py

#!/usr/bin/env python3
"""kis_trader/backtest/optuna_breakout.py — 돌파 Optuna (Grid add-on)."""
from __future__ import annotations

import json
import logging
import os
import time
from dataclasses import dataclass, field
from datetime import datetime
from typing import Any, Dict, List, Optional

import optuna
from optuna.samplers import RandomSampler, TPESampler

from database import TradeDB
from kis_trader.backtest import breakout_backtest_common as bbc
from kis_trader.backtest.optuna_search_space import breakout_grid_axis_keys, suggest_breakout_params
from kis_trader.backtest.optuna_breakout_tpe_space import (
    breakout_tpe_axis_keys,
    normalize_tpe_breakout_sl_mode,
    suggest_breakout_params_tpe,
)
from kis_trader.backtest.optuna_mode_combo import enrich_out_data_with_mode_combo
from kis_trader.backtest.optuna_common import (
    announce_optuna_json_path,
    build_optuna_result_tiers,
    pick_gated_apply_trial,
    release_shared_tick_store,
    set_optuna_trial_stability_attrs,
    stability_fields_from_trial_attrs,
)
from kis_trader.backtest.param_search_breakout import (
    _bo_fixed_defaults,
    _load_candles_for_search,
    _breakout_grids,
    _ui_to_engine_params,
    apply_params_to_db,
    evaluate_breakout_param_combo,
)
from kis_trader.backtest.param_search_cli_common import (
    apply_session_to_fixed,
    combo_passes_search_filters,
    format_session_hm,
)
from kis_trader.backtest.tail_param_search import _results_dir_for_write
from kis_trader.strategies.breakout import breakout_backtest_wants_tick_replay, breakout_entry_mode
from kis_trader.engine.indicator_cache import attach_indicator_caches_to_params
from kis_trader.backtest.breakout_tick_loader import load_breakout_ticks_by_code
from kis_trader.utils.env import get_env_bool

logger = logging.getLogger("param_search_optuna")

_FAIL_OBJECTIVE = -1e18


@dataclass
class BreakoutSearchContext:
    start: str
    end: str
    mode: str
    base_fixed: Dict[str, Any]
    codes_candles: Dict[str, List[Dict]]
    universe_by_slot: Optional[Dict[str, List[str]]]
    ticks_by_code: Any
    orderbook_by_code: Dict[str, Any]
    program_by_code: Dict[str, Any]
    log_verdict_by_code: Dict[str, Any]
    share_denom_by_code: Dict[str, float]
    fee_rate: float
    sell_tax: float
    slot_money: float
    max_stocks: int
    total_budget_krw: float
    period_days: int
    portfolio: Dict[str, Any]
    grid_keys: List[str]
    start_key: str
    end_key: str
    cache_holder: Dict[str, Any] = field(default_factory=dict)
    shared_tick_store: Any = None  # ws_ticks 공유메모리 핸들 (종료 시 unlink)
    tpe_sl_mode: str = "fixed"  # TPE 고정 손절모드(탐색 축 아님)


def prepare_breakout_search_context(
    start: str,
    end: str,
    mode: str,
    *,
    use_fallback_universe: bool = False,
    time_start_hm: Optional[int] = None,
    time_end_hm: Optional[int] = None,
    slot_money: Optional[float] = None,
    max_stocks: Optional[int] = None,
    total_budget_krw: Optional[float] = None,
    orderbook_filter: str = "off",
    history_source: Optional[str] = None,
    sl_mode: Optional[str] = None,
) -> Optional[BreakoutSearchContext]:
    grids = _breakout_grids()
    # tpe = 연속 Optuna 전용 (Grid 메뉴 미사용)
    if mode == "tpe":
        grid: Dict[str, Any] = {}
        logger.info(
            "📌 mode=tpe — 연속(float/int) 탐색 (Grid categorical 미사용, TPE 가 구간 축소)"
        )
    elif mode not in grids:
        logger.error("❌ 돌파 mode: %s (fast/coarse/fine/wide/full/tpe)", mode)
        return None
    else:
        grid = grids[mode]

    base_fixed = _bo_fixed_defaults()
    if mode == "tpe":
        base_fixed["skip_hts_scan_dupes"] = False
    apply_session_to_fixed(base_fixed, time_start_hm=time_start_hm, time_end_hm=time_end_hm)

    _ob_mode = (orderbook_filter or "off").strip().lower()
    if _ob_mode == "off":
        base_fixed["_orderbook_filter_enabled"] = False
    elif _ob_mode == "on":
        base_fixed["_orderbook_filter_enabled"] = True
    ob_filter_on = bool(base_fixed.get("_orderbook_filter_enabled")) or _ob_mode == "auto"
    logger.info(
        "📌 호가필터: %s (%s)",
        _ob_mode.upper(),
        "적용" if ob_filter_on else "스킵 — 코어 파라미터 순수 탐색",
    )

    db = TradeDB()
    try:
        from kis_trader.backtest.backtest_portfolio_common import load_portfolio_env_row
        env_row = load_portfolio_env_row(db)
    finally:
        db.close()

    fee_rate, sell_tax, slot_from_env = bbc.fee_and_slot_from_env(env_row)
    portfolio = bbc.resolve_breakout_portfolio_params(
        env_row, None,
        slot_money=slot_money if slot_money is not None else slot_from_env,
        max_stocks=max_stocks,
        total_budget_krw=total_budget_krw,
    )
    slot_money_f = float(portfolio["slot_money"])
    max_stocks_i = int(portfolio["max_stocks"])
    total_budget_f = float(portfolio["total_budget_krw"])
    period_days = max(
        1,
        (datetime.strptime(end, "%Y-%m-%d") - datetime.strptime(start, "%Y-%m-%d")).days + 1,
    )
    logger.info(
        f"💼 포트폴리오: 1회 {slot_money_f:,.0f}원 | 동시 {max_stocks_i}종 | "
        f"총한도 {total_budget_f:,.0f}원 | 매매 {format_session_hm(base_fixed)}"
    )
    logger.info("📌 진입 모드: %s", breakout_entry_mode())
    tpe_sl_mode = "fixed"
    if mode == "tpe":
        tpe_sl_mode = normalize_tpe_breakout_sl_mode(sl_mode)
        base_fixed["sl_mode"] = tpe_sl_mode
        logger.info(
            "📌 TPE 손절모드 고정: %s (탐색 축 아님 · atr_sl_*/sl_pct 중 해당만)",
            tpe_sl_mode,
        )

    from kis_trader.backtest.universe_history_source import (
        resolve_backtest_universe_history_source,
    )

    _hs = resolve_backtest_universe_history_source(history_source)
    base_fixed["_universe_history_source"] = _hs

    codes_candles = _load_candles_for_search(
        start, end, base_fixed.get("lookback_min", 1), base_fixed,
        history_source=_hs,
    )
    if not codes_candles:
        logger.error("❌ 캔들 데이터 없음")
        return None
    logger.info("✅ 데이터 로드: %s종목 (history=%s)", len(codes_candles), _hs)

    share_denom_by_code: Dict[str, float] = {}
    _share_db = TradeDB()
    try:
        from kis_trader.share.stock_share import load_share_denom_map
        share_denom_by_code = load_share_denom_map(_share_db, codes_candles.keys())
    finally:
        _share_db.close()

    start_key = (start.replace("-", "") + "0000") if start else "202601010000"
    end_key = (end.replace("-", "") + "2359") if end else "999912312359"
    start_ymd = start.replace("-", "") if start else ""
    end_ymd = end.replace("-", "") if end else ""

    ticks_by_code: Dict[str, Any] = {}
    engine_probe = _ui_to_engine_params(base_fixed)
    engine_probe["_orderbook_filter_enabled"] = base_fixed.get("_orderbook_filter_enabled")
    if breakout_backtest_wants_tick_replay(engine_probe):
        _tick_db = TradeDB()
        try:
            if _hs in ("ls", "ls_condition", "ls_ws"):
                from kis_trader.backtest.ls_history_loaders import load_ls_ticks_by_code

                ticks_by_code, tick_rows = load_ls_ticks_by_code(
                    _tick_db, start_key, end_key, set(codes_candles.keys()),
                )
                logger.info("✅ ls_ws_ticks %s건", f"{tick_rows:,}")
            else:
                ticks_by_code, tick_rows = load_breakout_ticks_by_code(
                    _tick_db, start_key, end_key, set(codes_candles.keys()),
                )
                logger.info("✅ ws_ticks %s건", f"{tick_rows:,}")
        finally:
            _tick_db.close()

    # ── ws_ticks 공유메모리 (Optuna, opt-in) — dict→numpy 컬럼 shared_memory 로 RAM 절감 ──
    # 끄려면 OPTUNA_PARAM_SEARCH_SHARED_TICKS=0. numpy/shm 미지원·빌드 실패 시 자동 폴백.
    shared_tick_store = None
    if get_env_bool("OPTUNA_PARAM_SEARCH_SHARED_TICKS", True) and ticks_by_code:
        from kis_trader.backtest.shared_ticks import build_shared_ticks_view
        _view, shared_tick_store = build_shared_ticks_view(ticks_by_code, enabled=True)
        if shared_tick_store is not None:
            import atexit as _atexit
            _atexit.register(shared_tick_store.unlink)  # 크래시 시 /dev/shm 누수 방지
            logger.info("📦 ws_ticks 공유메모리 ON (Optuna) — dict 사본 제거, RAM 절감")
            ticks_by_code = _view
            import gc as _gc
            _gc.collect()
            try:
                import ctypes as _ctypes
                _ctypes.CDLL("libc.so.6").malloc_trim(0)
            except Exception:
                pass

    # grid 는 상단에서 mode별 설정 (tpe=빈 dict). grids[mode] 재조회 금지.
    _ob_axes = ("max_spread_pct", "min_bid_ask_ratio", "ask_wall_max_qty")
    _ob_sweeping = any(len(set(grid.get(k) or [])) > 1 for k in _ob_axes)
    if ob_filter_on and _ob_sweeping:
        base_fixed["backtest_use_kiwoom_body_snapshot"] = True
        base_fixed["_backtest_use_kiwoom_body"] = True

    orderbook_by_code: Dict[str, Any] = {}
    program_by_code: Dict[str, Any] = {}
    log_verdict_by_code: Dict[str, Any] = {}
    _snap_db = TradeDB()
    try:
        from kis_trader.backtest.trigger_snapshot_loader import load_trigger_snapshots_by_code
        orderbook_by_code, program_by_code, trigger_snap_meta = load_trigger_snapshots_by_code(
            _snap_db, start_key, end_key, set(codes_candles.keys()),
            engine_params=engine_probe, strategy="BREAKOUT",
        )
        log_verdict_by_code = trigger_snap_meta.get("log_verdict_by_code") or {}
    finally:
        _snap_db.close()

    universe_by_slot = None
    fallback_sim_interval = 5
    if not use_fallback_universe and start_ymd and end_ymd:
        try:
            from kis_trader.backtest.breakout_backtest_common import resolve_breakout_universe

            history, src, n_bins, _scan_iv = resolve_breakout_universe(
                start_ymd, end_ymd, use_saved_history=True,
                history_source=_hs,
            )
            if history:
                universe_by_slot = history
                avg = sum(len(v) for v in history.values()) / max(1, n_bins)
                logger.info(
                    "✅ 유니버스: BREAKOUT 이력 src=%s | %s분봉 · 평균 %.1f종목",
                    src, n_bins, avg,
                )
        except Exception as exc:
            logger.debug("유니버스 이력 스킵: %s", exc)

    if universe_by_slot is None:
        from kis_trader.engine import scalping_engine as se
        universe_top_n = int(os.environ.get("UPDATE_UNIVERSE_TOP_N", "20"))
        universe_min_score = float(os.environ.get("UPDATE_UNIVERSE_MIN_SCORE", "4.0"))
        universe_by_slot = se.build_universe_simulation(
            codes_candles,
            top_n=universe_top_n,
            min_score=universe_min_score,
            scan_interval_min=fallback_sim_interval,
        )
        base_fixed["scan_interval_min"] = fallback_sim_interval
        logger.info("📌 유니버스: 시뮬 fallback (%d분)", fallback_sim_interval)
    else:
        base_fixed["scan_interval_min"] = 1

    cache_holder: Dict[str, Any] = {}
    attach_indicator_caches_to_params(cache_holder, codes_candles)

    return BreakoutSearchContext(
        start=start,
        end=end,
        mode=mode,
        base_fixed=base_fixed,
        codes_candles=codes_candles,
        universe_by_slot=universe_by_slot,
        ticks_by_code=ticks_by_code,
        orderbook_by_code=orderbook_by_code,
        program_by_code=program_by_code,
        log_verdict_by_code=log_verdict_by_code,
        share_denom_by_code=share_denom_by_code,
        fee_rate=fee_rate,
        sell_tax=sell_tax,
        slot_money=slot_money_f,
        max_stocks=max_stocks_i,
        total_budget_krw=total_budget_f,
        period_days=period_days,
        portfolio=portfolio,
        grid_keys=breakout_tpe_axis_keys(tpe_sl_mode) if mode == "tpe" else breakout_grid_axis_keys(mode),
        start_key=start_key,
        end_key=end_key,
        cache_holder=cache_holder,
        shared_tick_store=shared_tick_store,
        tpe_sl_mode=tpe_sl_mode,
    )


def _make_sampler(name: str, seed: Optional[int]):
    n = (name or "tpe").strip().lower()
    if n == "random":
        return RandomSampler(seed=seed)
    # multivariate TPE + 조건부 suggest 시 independent sampling 경고가 trial마다 폭주 → 억제
    return TPESampler(seed=seed, multivariate=True, warn_independent_sampling=False)


def run_breakout_optuna(
    ctx: BreakoutSearchContext,
    *,
    n_trials: int,
    storage_url: str,
    study_name: str,
    min_trades: int,
    min_win_rate: float,
    min_pf: float,
    sort_by: str = "pnl",
    sampler_name: str = "tpe",
    seed: Optional[int] = None,
    n_jobs: int = 1,
    show_progress: bool = True,
) -> optuna.Study:
    study = optuna.create_study(
        study_name=study_name,
        storage=storage_url,
        load_if_exists=True,
        direction="maximize",
        sampler=_make_sampler(sampler_name, seed),
    )

    def objective(trial: optuna.Trial) -> float:
        if ctx.mode == "tpe":
            combo = suggest_breakout_params_tpe(trial, sl_mode=ctx.tpe_sl_mode)
        else:
            combo = suggest_breakout_params(trial, ctx.mode)
        result = evaluate_breakout_param_combo(
            combo,
            base_fixed=ctx.base_fixed,
            grid_keys=ctx.grid_keys,
            codes_candles=ctx.codes_candles,
            min_trades=min_trades,
            min_win_rate=min_win_rate,
            min_pf=min_pf,
            universe_by_slot=ctx.universe_by_slot,
            slot_money=ctx.slot_money,
            max_stocks=ctx.max_stocks,
            total_budget_krw=ctx.total_budget_krw,
            fee_rate=ctx.fee_rate,
            sell_tax=ctx.sell_tax,
            period_days=ctx.period_days,
            cache_holder=ctx.cache_holder,
            ticks_by_code=ctx.ticks_by_code,
            orderbook_by_code=ctx.orderbook_by_code,
            program_by_code=ctx.program_by_code,
            log_verdict_by_code=ctx.log_verdict_by_code,
            share_denom_by_code=ctx.share_denom_by_code,
        )
        if result is None:
            trial.set_user_attr("gates_ok", False)
            return _FAIL_OBJECTIVE
        obj = float(result["win_rate"]) if sort_by == "win_rate" else float(result["total_pnl"])
        trial.set_user_attr("gates_ok", True)
        trial.set_user_attr("total_pnl", float(result["total_pnl"]))
        trial.set_user_attr("win_rate", float(result["win_rate"]))
        trial.set_user_attr("pf", float(result.get("pf") or 0))
        trial.set_user_attr("total_trades", int(result["total_trades"]))
        trial.set_user_attr("merged_json", json.dumps(result.get("merged_params") or {}, ensure_ascii=False))
        set_optuna_trial_stability_attrs(trial, result)
        return obj

    logger.info("🔬 Optuna BREAKOUT | study=%s | trials=%d", study_name, n_trials)
    t0 = time.time()
    try:
        study.optimize(objective, n_trials=n_trials, n_jobs=n_jobs, show_progress_bar=show_progress)
        elapsed = time.time() - t0

        passing: List[Dict[str, Any]] = []
        for trial in study.trials:
            if trial.state != optuna.trial.TrialState.COMPLETE:
                continue
            if not trial.user_attrs.get("gates_ok"):
                continue
            merged_raw = trial.user_attrs.get("merged_json") or "{}"
            try:
                merged = json.loads(merged_raw)
            except json.JSONDecodeError:
                merged = dict(trial.params)
            row = {
                "params": dict(trial.params),
                "merged_params": merged,
                "total_trades": int(trial.user_attrs.get("total_trades") or 0),
                "win_rate": float(trial.user_attrs.get("win_rate") or 0),
                "total_pnl": float(trial.user_attrs.get("total_pnl") or 0),
                "pf": float(trial.user_attrs.get("pf") or 0),
                "optuna_trial_number": trial.number,
            }
            row.update(stability_fields_from_trial_attrs(trial))
            passing.append(row)

        if sort_by == "win_rate":
            passing.sort(key=lambda r: (-r["win_rate"], -r["total_pnl"]))
        else:
            passing.sort(key=lambda r: (-r["total_pnl"], -r["win_rate"]))
        tiers = build_optuna_result_tiers(passing, sort_by=sort_by)

        hints: Dict[str, str] = {}
        out_data = {
            "engine": "optuna",
            "strategy": "breakout",
            "mode": ctx.mode,
            "start": ctx.start,
            "end": ctx.end,
            "slot_money": int(ctx.slot_money),
            "max_stocks": ctx.max_stocks,
            "total_budget_krw": int(ctx.total_budget_krw),
            "backtest_days": ctx.period_days,
            "min_trades": min_trades,
            "min_win_rate": min_win_rate,
            "min_pf": min_pf,
            "sort_by": sort_by,
            "grid_keys": ctx.grid_keys,
            "grid_axis_hints": {k: hints[k] for k in ctx.grid_keys if k in hints},
            "optuna_study_name": study_name,
            "optuna_storage": storage_url,
            "optuna_n_trials_requested": n_trials,
            "optuna_trials_completed": len(study.trials),
            "optuna_best_value": study.best_value if study.best_trial else None,
            "optuna_best_trial_number": study.best_trial.number if study.best_trial else None,
            "elapsed_sec": round(elapsed, 1),
            **tiers,
        }

        ts = datetime.now().strftime("%Y%m%d_%H%M%S")
        out_path = os.path.join(_results_dir_for_write(), f"optuna_breakout_{ctx.mode}_{ts}.json")
        with open(out_path, "w", encoding="utf-8") as f:
            json.dump(out_data, f, indent=2, ensure_ascii=False)
        announce_optuna_json_path(
            out_path, strategy="breakout", mode=ctx.mode, note="중간저장(mode 전)", log=logger,
        )

        def _eval_mode(combo: Dict[str, Any]) -> Optional[Dict[str, Any]]:
            return evaluate_breakout_param_combo(
                combo,
                base_fixed=ctx.base_fixed,
                grid_keys=ctx.grid_keys,
                codes_candles=ctx.codes_candles,
                min_trades=1,
                min_win_rate=0.0,
                min_pf=0.0,
                universe_by_slot=ctx.universe_by_slot,
                slot_money=ctx.slot_money,
                max_stocks=ctx.max_stocks,
                total_budget_krw=ctx.total_budget_krw,
                fee_rate=ctx.fee_rate,
                sell_tax=ctx.sell_tax,
                period_days=ctx.period_days,
                cache_holder=ctx.cache_holder,
                ticks_by_code=ctx.ticks_by_code,
                orderbook_by_code=ctx.orderbook_by_code,
                program_by_code=ctx.program_by_code,
                log_verdict_by_code=ctx.log_verdict_by_code,
                share_denom_by_code=ctx.share_denom_by_code,
                include_trades=True,
            )

        def _save_partial(_data: Dict[str, Any]) -> None:
            with open(out_path, "w", encoding="utf-8") as f:
                json.dump(_data, f, indent=2, ensure_ascii=False)
            announce_optuna_json_path(
                out_path, strategy="breakout", mode=ctx.mode, note="mode_combo params 저장(실측 전)", log=logger,
            )

        enrich_out_data_with_mode_combo(
            out_data,
            evaluate_fn=_eval_mode,
            grid_keys=ctx.grid_keys,
            log=logger,
            on_partial_save=_save_partial,
        )
        with open(out_path, "w", encoding="utf-8") as f:
            json.dump(out_data, f, indent=2, ensure_ascii=False)
        announce_optuna_json_path(
            out_path, strategy="breakout", mode=ctx.mode, note="최종 JSON", log=logger,
        )
        study._kis_export_path = out_path  # type: ignore[attr-defined]
        return study
    finally:
        release_shared_tick_store(ctx, log=logger)


def apply_best_breakout_trial(study: optuna.Study) -> bool:
    trial = pick_gated_apply_trial(study, sort_by="pnl", fail_objective=_FAIL_OBJECTIVE)
    if trial is None:
        logger.warning(
            "⚠️ 사후게이트(results_gated) 통과 trial 없음 — DB 미적용"
        )
        return False
    pnl = float(trial.user_attrs.get("total_pnl") or 0)
    if pnl <= 0:
        logger.warning("⚠️ gated trial 총손익 ≤ 0 — DB 미적용")
        return False
    merged = json.loads(trial.user_attrs.get("merged_json") or "{}")
    apply_params_to_db(merged)
    logger.info("🚀 [Optuna apply-best] breakout gated trial #%d → env_config", trial.number)
    try:
        from kis_trader.backtest.optuna_daily_trail_recommend import (
            apply_daily_trail_recommend_from_optuna_json,
        )
        apply_daily_trail_recommend_from_optuna_json(
            getattr(study, "_kis_export_path", None),
            strategy="breakout",
            log=logger,
        )
    except Exception as exc:
        logger.warning("⚠️ 다단트레일 추천 반영 스킵: %s", exc)
    return True