Files
trader-bot/backend/app/services/market_data.py
tika 4df8d53b1a feat: trading bot MVP — ICT Order Block + Liquidity Sweep strategy
Full-stack trading bot with:
- FastAPI backend with ICT strategy (Order Block + Liquidity Sweep detection)
- Backtester engine with rolling window, spread simulation, and performance metrics
- Hybrid market data service (yfinance + TwelveData with rate limiting + SQLite cache)
- Simulated exchange for paper trading
- React/TypeScript frontend with TradingView lightweight-charts v5
- Live dashboard with candlestick chart, OHLC legend, trade markers
- Backtest page with configurable parameters, equity curve, and trade table
- WebSocket support for real-time updates
- Bot runner with asyncio loop for automated trading

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-24 23:25:51 +01:00

265 lines
10 KiB
Python

"""
MarketDataService — source de données hybride avec cache DB.
Stratégie de fetch pour une période [start, end] demandée :
1. DB d'abord → on récupère ce qu'on a déjà, on ne refetch jamais ce qui existe
2. Gaps récents → yfinance (dans ses limites temporelles)
3. Gaps historiques → TwelveData (pour tout ce que yfinance ne peut pas couvrir)
4. Tout est stocké → les prochaines requêtes seront servies depuis la DB
Exemple (M1, 10 derniers jours demandés) :
- DB : déjà ce qu'on a en cache
- yfinance : J-7 → maintenant (limite M1 = 7 jours)
- TwelveData : J-10 → J-7 (historique au-delà de yfinance)
"""
import logging
from datetime import datetime, timedelta
from typing import Optional
import pandas as pd
from sqlalchemy import and_, select, text
from sqlalchemy.ext.asyncio import AsyncSession
from app.models.candle import Candle
from app.services.data_providers.constants import GRANULARITY_MINUTES
from app.services.data_providers.twelvedata_provider import TwelveDataProvider
from app.services.data_providers.yfinance_provider import YFinanceProvider
logger = logging.getLogger(__name__)
# Facteur pour compenser weekends + jours fériés dans le calcul de la fenêtre
TRADING_DAYS_FACTOR = 1.5
class MarketDataService:
def __init__(self, db: AsyncSession) -> None:
self._db = db
self._yf = YFinanceProvider()
self._td = TwelveDataProvider()
# ── API publique ──────────────────────────────────────────────────────────
async def get_candles(
self,
instrument: str,
granularity: str,
count: int = 200,
start: Optional[datetime] = None,
end: Optional[datetime] = None,
) -> pd.DataFrame:
"""
Retourne jusqu'à `count` bougies pour instrument/granularity.
Si start/end fournis, ils définissent la plage exacte.
Processus :
1. Calcul de la fenêtre temporelle nécessaire
2. Détection et comblement des gaps (yfinance + TwelveData)
3. Lecture depuis DB et retour
"""
if end is None:
end = datetime.utcnow()
if start is None:
minutes = GRANULARITY_MINUTES.get(granularity, 60)
start = end - timedelta(minutes=int(minutes * count * TRADING_DAYS_FACTOR))
await self._fill_gaps(instrument, granularity, start, end)
return await self._db_fetch(instrument, granularity, start, end, limit=count)
async def get_latest_price(self, instrument: str) -> Optional[float]:
"""Retourne le dernier close connu (DB ou yfinance M1)."""
stmt = (
select(Candle.close)
.where(Candle.instrument == instrument)
.order_by(Candle.time.desc())
.limit(1)
)
result = await self._db.execute(stmt)
price = result.scalar_one_or_none()
if price:
return float(price)
df = await self.get_candles(instrument, "M1", count=2)
return float(df.iloc[-1]["close"]) if not df.empty else None
# ── Logique de détection et comblement des gaps ───────────────────────────
async def _fill_gaps(
self,
instrument: str,
granularity: str,
start: datetime,
end: datetime,
) -> None:
gaps = await self._find_gaps(instrument, granularity, start, end)
for gap_start, gap_end in gaps:
await self._fetch_and_store_gap(instrument, granularity, gap_start, gap_end)
async def _find_gaps(
self,
instrument: str,
granularity: str,
start: datetime,
end: datetime,
) -> list[tuple[datetime, datetime]]:
"""
Retourne la liste des (gap_start, gap_end) manquants en DB.
Logique :
- Si rien en DB pour la plage → un seul gap = (start, end)
- Sinon → combler avant le plus ancien et/ou après le plus récent
"""
stmt = (
select(Candle.time)
.where(
and_(
Candle.instrument == instrument,
Candle.granularity == granularity,
Candle.time >= start,
Candle.time <= end,
)
)
.order_by(Candle.time)
)
result = await self._db.execute(stmt)
times = [r[0] for r in result.fetchall()]
if not times:
return [(start, end)]
gaps: list[tuple[datetime, datetime]] = []
interval = timedelta(minutes=GRANULARITY_MINUTES.get(granularity, 60))
oldest, newest = times[0], times[-1]
# Gap avant : demande de données antérieures à ce qu'on a
if start < oldest - interval:
gaps.append((start, oldest))
# Gap après : demande de données plus récentes que ce qu'on a
freshness_threshold = interval * 2
if end > newest + freshness_threshold:
gaps.append((newest, end))
return gaps
async def _fetch_and_store_gap(
self,
instrument: str,
granularity: str,
gap_start: datetime,
gap_end: datetime,
) -> None:
"""
Fetche un gap :
1. yfinance pour la partie récente (dans ses limites)
2. TwelveData en fallback si yfinance échoue, ou pour la partie historique
"""
yf_cutoff = self._yf.yf_cutoff(granularity)
yf_covered = False
# ── yfinance : partie récente du gap ─────────────────────────────────
if yf_cutoff is not None:
yf_start = max(gap_start, yf_cutoff)
if yf_start < gap_end:
df_yf = await self._yf.fetch(instrument, granularity, yf_start, gap_end)
if not df_yf.empty:
await self._store(df_yf, instrument, granularity)
yf_covered = True
# ── TwelveData : historique + fallback si yfinance indisponible ───────
if self._td.is_configured():
# Partie historique (avant la limite yfinance)
td_end = yf_cutoff if (yf_cutoff and gap_start < yf_cutoff) else None
if td_end and gap_start < td_end:
df_td = await self._td.fetch(instrument, granularity, gap_start, td_end)
if not df_td.empty:
await self._store(df_td, instrument, granularity)
# Fallback pour la partie récente si yfinance n'a rien retourné
if not yf_covered:
yf_start = max(gap_start, yf_cutoff) if yf_cutoff else gap_start
if yf_start < gap_end:
logger.info(
"yfinance indisponible — fallback TwelveData pour %s %s [%s%s]",
instrument, granularity,
yf_start.strftime("%Y-%m-%d"), gap_end.strftime("%Y-%m-%d"),
)
df_td2 = await self._td.fetch(instrument, granularity, yf_start, gap_end)
if not df_td2.empty:
await self._store(df_td2, instrument, granularity)
elif not yf_covered:
logger.warning(
"Gap [%s%s] pour %s %s"
"TWELVEDATA_API_KEY manquante et yfinance indisponible.",
gap_start.strftime("%Y-%m-%d"),
gap_end.strftime("%Y-%m-%d"),
instrument,
granularity,
)
# ── DB helpers ────────────────────────────────────────────────────────────
async def _db_fetch(
self,
instrument: str,
granularity: str,
start: datetime,
end: datetime,
limit: int = 5000,
) -> pd.DataFrame:
stmt = (
select(Candle)
.where(
and_(
Candle.instrument == instrument,
Candle.granularity == granularity,
Candle.time >= start,
Candle.time <= end,
)
)
.order_by(Candle.time.desc())
.limit(limit)
)
result = await self._db.execute(stmt)
rows = result.scalars().all()
if not rows:
return pd.DataFrame(columns=["time", "open", "high", "low", "close", "volume"])
df = pd.DataFrame(
[{"time": r.time, "open": r.open, "high": r.high,
"low": r.low, "close": r.close, "volume": r.volume}
for r in rows]
)
return df.sort_values("time").reset_index(drop=True)
async def _store(self, df: pd.DataFrame, instrument: str, granularity: str) -> None:
"""
Insère les bougies en DB avec INSERT OR IGNORE.
Les bougies déjà présentes (même instrument+granularity+time) ne sont jamais modifiées.
"""
if df.empty:
return
for _, row in df.iterrows():
await self._db.execute(
text(
"INSERT OR IGNORE INTO candles "
"(instrument, granularity, time, open, high, low, close, volume, complete) "
"VALUES (:instrument, :granularity, :time, :open, :high, :low, :close, :volume, 1)"
),
{
"instrument": instrument,
"granularity": granularity,
"time": pd.Timestamp(row["time"]).to_pydatetime().replace(tzinfo=None),
"open": float(row["open"]),
"high": float(row["high"]),
"low": float(row["low"]),
"close": float(row["close"]),
"volume": int(row.get("volume", 0)),
},
)
await self._db.commit()