Files
OpenFin/backend/routers/wavelet.py
2026-07-20 21:43:35 +02:00

256 lines
12 KiB
Python

"""
Wavelet decomposition endpoints — adapted from InstrumentSimulator's /api/wavelet/analyze
and /api/wavelet/rolling. That project fetches history from a Postgres-cached date-range
store; we don't have that, so history comes from services.data_fetcher.get_historical()
(yfinance period/interval strings by default, or explicit start/end dates when a custom
range is requested). The wavelet math itself (services.wavelet_engine) is an unmodified
port.
"""
from datetime import date, datetime, timedelta
from typing import Any, Dict, List, Optional
from fastapi import APIRouter, HTTPException, Query
from pydantic import BaseModel
router = APIRouter(prefix="/api/wavelet", tags=["wavelet"])
# Approximate calendar-day span of each yfinance period string, used to pick a fetch
# window large enough to cover the requested causal (rolling) output range plus the
# lookback padding needed to fill the first walk-forward window.
_PERIOD_TO_DAYS = {
"5d": 5, "1mo": 30, "3mo": 90, "6mo": 182,
"1y": 365, "2y": 730, "5y": 1825, "10y": 3650, "max": 3650,
}
def _fetch_history(symbol: str, period: str, interval: str = "1d", start: Optional[str] = None, end: Optional[str] = None):
from services.data_fetcher import get_historical
hist = get_historical(symbol, period=period, interval=interval, start=start, end=end)
values = [h["close"] for h in hist]
dates = [h["date"] for h in hist]
return values, dates
def _fetch_padded_history(
symbol: str, lookback: int, period: str = "1y",
start_date: Optional[str] = None, end_date: Optional[str] = None,
future_padding_days: int = 0,
):
"""Fetch enough history to cover the requested causal-output range plus a
`lookback`-sized warmup window before it (mirrors main.py's fetch_start padding).
Returns (values, dates, cutoff) — `cutoff` is the first date that belongs in the
causal OUTPUT range; everything in `dates`/`values` before it is lookback padding.
Two modes:
- `period` (default): output range = the last `period` worth of days up to today,
cutoff computed backward from now. The fetch already reaches "today" regardless
of `future_padding_days` (yfinance can't return data past now), so that argument
is a no-op here — the reliability endpoint's per-turning-point "not enough real
future data yet" guard is what actually limits which recent points are testable.
- `start_date`/`end_date`: output range = that explicit window (e.g. a custom
backtest over 2013-2019) — fetches from `start_date` minus the lookback padding
through `end_date` (or through today if `end_date` is omitted). Unlike `period`
mode, `end_date` is an arbitrary cutoff with real data available past it, so
`future_padding_days` explicitly extends the fetch past it — needed for the
reliability endpoint to have real hindsight data for turning points near the end
of a custom range.
"""
pad_days = int(lookback * 1.6) + 15
if start_date:
fetch_start = (date.fromisoformat(start_date) - timedelta(days=pad_days)).isoformat()
fetch_end = (date.fromisoformat(end_date) + timedelta(days=future_padding_days)).isoformat() if end_date and future_padding_days else end_date
values, dates = _fetch_history(symbol, period, start=fetch_start, end=fetch_end)
return values, dates, start_date
out_days = _PERIOD_TO_DAYS.get(period, 365)
total_days = out_days + pad_days
fetch_period = next(
(p for p, d in sorted(_PERIOD_TO_DAYS.items(), key=lambda kv: kv[1]) if d >= total_days),
"10y",
)
values, dates = _fetch_history(symbol, fetch_period)
cutoff = (datetime.utcnow() - timedelta(days=out_days)).date().isoformat()
return values, dates, cutoff
@router.get("/analyze")
def wavelet_analyze(
symbol: str = Query("SPY"),
period: str = Query("1y", description="yfinance period: 5d,1mo,3mo,6mo,1y,2y,5y,10y,max — ignored if start_date is set"),
levels: int = Query(4, ge=2, le=6),
wavelet: str = Query("gmw", description="gmw, morlet or bump"),
window_size: int = Query(0, description="0 = single window over the whole range"),
method: str = Query("cwt", description="cwt (default) or ssq"),
start_date: Optional[str] = Query(None, description="ISO date (YYYY-MM-DD) — overrides `period` with an explicit custom range"),
end_date: Optional[str] = Query(None, description="ISO date (YYYY-MM-DD), only used alongside start_date; omit for 'through today'"),
):
from services.wavelet_engine import windowed_band_decompose, band_decompose_ssq
values, dates = _fetch_history(symbol, period, start=start_date, end=end_date)
if len(values) < 32:
raise HTTPException(400, "Historique insuffisant pour une analyse ondelette (32 points minimum).")
try:
if method == "ssq":
# No windowed/chunked variant for ssq in the source project — analyze the
# whole fetched range in one pass (matches window_size=0 semantics).
return band_decompose_ssq(values, dates, num_levels=levels, wavelet=wavelet)
return windowed_band_decompose(values, dates, window_size=window_size, num_levels=levels, wavelet=wavelet)
except ValueError as exc:
raise HTTPException(400, str(exc)) from exc
@router.get("/rolling")
def wavelet_rolling(
symbol: str = Query("SPY"),
period: str = Query("1y", description="how much of the causal output range to return — ignored if start_date is set"),
lookback: int = Query(260, ge=32),
levels: int = Query(4, ge=2, le=6),
wavelet: str = Query("gmw"),
step: int = Query(1, ge=1),
method: str = Query("cwt", description="cwt (default) or ssq"),
start_date: Optional[str] = Query(None, description="ISO date (YYYY-MM-DD) — overrides `period`; the causal output starts here"),
end_date: Optional[str] = Query(None, description="ISO date (YYYY-MM-DD), only used alongside start_date; omit for 'through today'"),
):
"""Walk-forward version of /analyze: band values are computed day by day from a
trailing `lookback`-point window only, so a trade simulation built on this never
sees data past its own decision date."""
from services.wavelet_engine import rolling_causal_bands, rolling_causal_bands_ssq
values, dates, cutoff = _fetch_padded_history(symbol, lookback, period, start_date, end_date)
if len(values) < lookback + 32:
raise HTTPException(400, "Historique insuffisant pour une analyse ondelette (32 points minimum).")
start_idx = next((i for i, d in enumerate(dates) if d[:10] >= cutoff), None)
if start_idx is None:
raise HTTPException(400, "Pas de donnees dans la plage de trading demandee.")
decomposer = rolling_causal_bands_ssq if method == "ssq" else rolling_causal_bands
try:
result = decomposer(
values, dates,
start_idx=start_idx, lookback=lookback,
num_levels=levels, wavelet=wavelet, step=step,
)
except ValueError as exc:
raise HTTPException(400, str(exc)) from exc
if not result["dates"]:
raise HTTPException(
400,
f"Pas assez d'historique avant {cutoff} pour remplir une fenetre glissante de {lookback} points.",
)
return result
@router.get("/reliability")
def wavelet_reliability_endpoint(
symbol: str = Query("SPY"),
period: str = Query("1y", description="how much of the causal output range to scan for turning points — ignored if start_date is set"),
lookback: int = Query(260, ge=32),
levels: int = Query(4, ge=2, le=6),
wavelet: str = Query("gmw"),
step: int = Query(1, ge=1),
method: str = Query("cwt", description="cwt (default) or ssq"),
smooth_days: int = Query(3, ge=1, le=10, description="lag used to smooth the slope before flagging a sign-change as a turning point"),
tolerance_pct: float = Query(0.10, ge=0, le=0.5, description="date-matching tolerance as a fraction of each band's own average cycle length, applied both sides"),
min_confirm_horizon: int = Query(3, ge=1, le=30, description="floor on the hindsight horizon in days, in case a band's measured cycle length comes out very short"),
max_future_padding: int = Query(60, ge=10, le=180, description="extra real days fetched beyond the requested range, to cover the slowest band's own (data-driven) confirm horizon"),
start_date: Optional[str] = Query(None, description="ISO date (YYYY-MM-DD) — overrides `period`; the causal output starts here"),
end_date: Optional[str] = Query(None, description="ISO date (YYYY-MM-DD), only used alongside start_date; omit for 'through today'"),
):
"""For every reversal a live (causal, walk-forward) decomposition would have flagged,
checks whether redoing the decomposition later still shows the same reversal — a
per-band reliability score. The hindsight horizon is dynamic per band (each band's own
historical average peak-to-peak cycle length, not one fixed day count for every band —
a fixed horizon is meaningless across bands with wildly different natural periods)."""
from services.wavelet_engine import wavelet_reliability
# The horizon is now computed per band inside wavelet_reliability (from each band's own
# measured cycle length), so we don't know it in advance here — pad generously enough to
# cover even a slow band's cycle instead.
values, dates, cutoff = _fetch_padded_history(symbol, lookback, period, start_date, end_date, future_padding_days=max_future_padding)
if len(values) < lookback + max_future_padding + 32:
raise HTTPException(400, "Historique insuffisant pour un test de fiabilité (32 points minimum au-delà de la fenêtre + marge).")
start_idx = next((i for i, d in enumerate(dates) if d[:10] >= cutoff), None)
if start_idx is None:
raise HTTPException(400, "Pas de donnees dans la plage de trading demandee.")
try:
result = wavelet_reliability(
values, dates,
start_idx=start_idx, lookback=lookback,
num_levels=levels, wavelet=wavelet, method=method, step=step,
smooth_days=smooth_days, tolerance_pct=tolerance_pct, min_confirm_horizon=min_confirm_horizon,
)
except ValueError as exc:
raise HTTPException(400, str(exc)) from exc
return result
# ── Saved simulation/optimization runs ────────────────────────────────────────
class SimulationCreate(BaseModel):
name: str
form: Dict[str, Any] = {}
results: List[Dict[str, Any]] = []
excluded_instruments: List[str] = []
class SimulationUpdate(BaseModel):
name: Optional[str] = None
form: Optional[Dict[str, Any]] = None
results: Optional[List[Dict[str, Any]]] = None
append_results: Optional[List[Dict[str, Any]]] = None
excluded_instruments: Optional[List[str]] = None
@router.get("/simulations")
def list_simulations():
from services.database import get_wavelet_simulations
return get_wavelet_simulations()
@router.post("/simulations")
def create_simulation(payload: SimulationCreate):
from services.database import save_wavelet_simulation
return save_wavelet_simulation(payload.name, payload.form, payload.results, payload.excluded_instruments)
@router.get("/simulations/{sim_id}")
def get_simulation(sim_id: str):
from services.database import get_wavelet_simulation
sim = get_wavelet_simulation(sim_id)
if not sim:
raise HTTPException(404, "Simulation introuvable.")
return sim
@router.patch("/simulations/{sim_id}")
def patch_simulation(sim_id: str, payload: SimulationUpdate):
from services.database import update_wavelet_simulation
sim = update_wavelet_simulation(
sim_id, name=payload.name, form=payload.form, results=payload.results,
append_results=payload.append_results, excluded_instruments=payload.excluded_instruments,
)
if not sim:
raise HTTPException(404, "Simulation introuvable.")
return sim
@router.delete("/simulations/{sim_id}")
def remove_simulation(sim_id: str):
from services.database import delete_wavelet_simulation
if not delete_wavelet_simulation(sim_id):
raise HTTPException(404, "Simulation introuvable.")
return {"deleted": sim_id}
# ── Watchlist signal scan (populated by the auto-cycle, see services.wavelet_signals) ──
@router.get("/watchlist-signals")
def watchlist_signals():
from services.database import get_latest_wavelet_signals
return {"signals": get_latest_wavelet_signals()}