kis_bot/kis_trader/backtest/optuna_momentum.py

#!/usr/bin/env python3
"""kis_trader/backtest/optuna_momentum.py — 모멘텀 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 momentum_backtest_common as mbc
from kis_trader.backtest import scalping_backtest_common as sbc
from kis_trader.backtest.optuna_search_space import momentum_grid_axis_keys, suggest_momentum_params
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, release_shared_tick_store
from kis_trader.backtest.param_search_cli_common import (
    apply_session_to_fixed,
    combo_passes_search_filters,
    format_session_hm,
)
from kis_trader.backtest.param_search_momentum import (
    MOMENTUM_GRID_AXIS_HINTS_KO,
    _load_candles_for_search,
    _mom_fixed_defaults,
    _momentum_grids,
    apply_params_to_db,
    evaluate_momentum_param_combo,
)
from kis_trader.backtest.tail_param_search import _results_dir_for_write
from kis_trader.engine import momentum_engine as me
from kis_trader.engine.indicator_cache import attach_indicator_caches_to_params
from kis_trader.utils.env import get_env_bool, get_env_float

logger = logging.getLogger("param_search_optuna")

_FAIL_OBJECTIVE = -1e18


@dataclass
class MomentumSearchContext:
    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]
    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)


def prepare_momentum_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",
) -> Optional[MomentumSearchContext]:
    grids = _momentum_grids()
    if mode not in grids:
        logger.error("❌ 모멘텀 mode: %s (fast/exit/rr/coarse/fine/wide/full)", mode)
        return None

    base_fixed = _mom_fixed_defaults()
    apply_session_to_fixed(base_fixed, time_start_hm=time_start_hm, time_end_hm=time_end_hm)

    from kis_trader.engine.momentum_tick_replay import (
        momentum_backtest_use_tick_entry as _te,
        momentum_backtest_use_tick_exit as _tx,
    )
    base_fixed["backtest_use_tick_entry"] = _te(None)
    base_fixed["backtest_use_tick_exit"] = _tx(None)

    _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 = sbc.fee_and_slot_from_env(env_row, strategy="MOMENTUM")
    portfolio = sbc.resolve_scalp_portfolio_params(
        env_row, None, strategy="MOMENTUM",
        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)}"
    )

    codes_candles = _load_candles_for_search(start, end, base_fixed.get("rsi_period", 3))
    if not codes_candles:
        logger.error("❌ 캔들 데이터 없음")
        return None
    logger.info("✅ 데이터 로드: %s종목", len(codes_candles))

    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 ""

    universe_by_slot = None
    fallback_sim_interval = 5
    if not use_fallback_universe and start_ymd and end_ymd:
        try:
            from kis_trader.backtest.momentum_backtest_common import resolve_momentum_universe
            history, src, n_bins, _scan_iv, timing = resolve_momentum_universe(
                start_ymd, end_ymd, use_saved_history=True, strategy_id="MOMENTUM",
            )
            if history:
                universe_by_slot = history
                avg = sum(len(v) for v in history.values()) / max(1, n_bins)
                logger.info(
                    "✅ 유니버스: MOMENTUM 이력 | %s분봉 · 평균 %.1f종목", n_bins, avg,
                )
        except Exception as exc:
            logger.debug("유니버스 이력 스킵: %s", exc)

    if universe_by_slot is None:
        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 = me.build_universe_simulation_momentum(
            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: 모멘텀 시뮬 fallback (%d분)", fallback_sim_interval)
    else:
        base_fixed["scan_interval_min"] = 1

    # DB 전일봉 없으면 키움 REST 1회/종목 → 메모리 prepend (실매 갭보정 정합, DB 미기록)
    try:
        from kis_trader.backtest.momentum_backtest_common import (
            inject_momentum_rest_warmup_memory,
        )
        _rw = inject_momentum_rest_warmup_memory(
            codes_candles,
            start_key,
            universe_by_slot=universe_by_slot,
        )
        if int(_rw.get("ok") or 0) > 0 or int(_rw.get("need") or 0) > 0:
            logger.info(
                "📡 REST 웜업: need=%s ok=%s fail=%s bars=%s",
                _rw.get("need"), _rw.get("ok"), _rw.get("fail"), _rw.get("bars"),
            )
    except Exception as exc:
        logger.warning("⚠️ REST 웜업 스킵: %s", exc)

    grid = grids[mode]
    _ob_axes = ("max_spread_pct", "min_bid_ask_ratio", "ask_max_mult")
    _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] = {}
    ticks_by_code: Dict[str, Any] = {}
    _snap_db = TradeDB()
    try:
        from kis_trader.backtest.trigger_snapshot_loader import (
            backtest_needs_trigger_snapshot_load,
            load_trigger_snapshots_by_code,
        )
        from kis_trader.engine.momentum_tick_replay import (
            momentum_backtest_use_tick_entry,
            momentum_backtest_use_tick_exit,
        )
        if backtest_needs_trigger_snapshot_load(base_fixed, strategy="MOMENTUM"):
            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=base_fixed, strategy="MOMENTUM",
            )
            log_verdict_by_code = trigger_snap_meta.get("log_verdict_by_code") or {}
        if momentum_backtest_use_tick_exit(base_fixed) or momentum_backtest_use_tick_entry(base_fixed):
            from kis_trader.backtest.momentum_tick_loader import load_momentum_ticks_by_code
            ticks_by_code, tick_rows = load_momentum_ticks_by_code(
                _snap_db, start_key, end_key, set(codes_candles.keys()),
            )
            logger.info("✅ ws_ticks %s건", f"{tick_rows:,}")
    finally:
        _snap_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

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

    return MomentumSearchContext(
        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,
        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=momentum_grid_axis_keys(mode),
        start_key=start_key,
        end_key=end_key,
        cache_holder=cache_holder,
        shared_tick_store=shared_tick_store,
    )


def _make_sampler(name: str, seed: Optional[int]):
    n = (name or "tpe").strip().lower()
    if n == "random":
        return RandomSampler(seed=seed)
    return TPESampler(seed=seed, multivariate=True)


def _momentum_objective_value(result: Dict[str, Any], sort_by: str) -> float:
    pnl = float(result["total_pnl"])
    if sort_by == "score":
        mdd_floor = get_env_float("MOMENTUM_SCORE_MDD_FLOOR", 10000.0)
        mdd = float(result.get("mdd") or 0)
        return pnl / max(mdd, mdd_floor)
    if sort_by == "win_rate":
        return float(result["win_rate"])
    return pnl


def run_momentum_optuna(
    ctx: MomentumSearchContext,
    *,
    n_trials: int,
    storage_url: str,
    study_name: str,
    min_trades: int,
    min_win_rate: float,
    min_pf: float,
    sort_by: str = "score",
    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:
        combo = suggest_momentum_params(trial, ctx.mode)
        result = evaluate_momentum_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,
            start_key=ctx.start_key,
            end_key=ctx.end_key,
        )
        if result is None:
            trial.set_user_attr("gates_ok", False)
            return _FAIL_OBJECTIVE
        obj = _momentum_objective_value(result, sort_by)
        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("mdd", float(result.get("mdd") or 0))
        trial.set_user_attr("score", float(obj if sort_by == "score" else _momentum_objective_value(result, "score")))
        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))
        return float(obj)

    logger.info(
        "🔬 Optuna MOMENTUM | study=%s | trials=%d | sort=%s",
        study_name, n_trials, sort_by,
    )
    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),
                "mdd": float(trial.user_attrs.get("mdd") or 0),
                "score": float(trial.user_attrs.get("score") or 0),
                "optuna_trial_number": trial.number,
            }
            passing.append(row)

        if sort_by == "score":
            passing.sort(key=lambda r: (-r["score"], -r["total_pnl"], -r["win_rate"]))
        elif 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"]))

        profitable = [r for r in passing if r["total_pnl"] > 0]
        if profitable:
            passing = profitable

        out_data = {
            "engine": "optuna",
            "strategy": "momentum",
            "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: MOMENTUM_GRID_AXIS_HINTS_KO[k] for k in ctx.grid_keys if k in MOMENTUM_GRID_AXIS_HINTS_KO},
            "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),
            "results": passing[:5000],
        }

        ts = datetime.now().strftime("%Y%m%d_%H%M%S")
        out_path = os.path.join(_results_dir_for_write(), f"optuna_momentum_{ctx.mode}_{ts}.json")
        # 최빈 실측 전에 먼저 저장·경로 고지 (실측이 길어도 바로 파일 열 수 있게)
        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="momentum", mode=ctx.mode, note="중간저장(mode 전)", log=logger,
        )

        def _eval_mode(combo: Dict[str, Any]) -> Optional[Dict[str, Any]]:
            # 최빈 Frankenstein 실측 — 게이트는 느슨하게(리포트용)
            # ※ shared_tick_store 가 아직 살아 있어야 함 (optimize 직후 unlink 금지)
            return evaluate_momentum_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,
                start_key=ctx.start_key,
                end_key=ctx.end_key,
            )

        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="momentum", 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="momentum", mode=ctx.mode, note="최종 JSON", log=logger,
        )
        study._kis_export_path = out_path  # type: ignore[attr-defined]
        return study
    finally:
        # mode_combo 실측이 ticks 공유메모리 뷰를 쓰므로, 여기서 해제 (optimize 직후 X)
        release_shared_tick_store(ctx, log=logger)


def apply_best_momentum_trial(study: optuna.Study) -> bool:
    if not study.best_trial or study.best_value <= _FAIL_OBJECTIVE + 1:
        logger.warning("⚠️ 적용할 best trial 없음")
        return False
    pnl = float(study.best_trial.user_attrs.get("total_pnl") or 0)
    if pnl <= 0:
        logger.warning("⚠️ Best trial 총손익 ≤ 0 — DB 미적용")
        return False
    merged_raw = study.best_trial.user_attrs.get("merged_json") or "{}"
    merged = json.loads(merged_raw)
    apply_params_to_db(merged)
    logger.info("🚀 [Optuna apply-best] momentum trial #%d → env_config", study.best_trial.number)
    return True