Initial strategy32 research and live runtime

This commit is contained in:
2026-03-16 20:18:41 -07:00
commit c165a9add7
42 changed files with 10750 additions and 0 deletions

View File

@@ -0,0 +1,296 @@
from __future__ import annotations
import json
import os
import sys
from dataclasses import asdict
from pathlib import Path
import pandas as pd
PACKAGE_PARENT = Path(__file__).resolve().parents[2]
if str(PACKAGE_PARENT) not in sys.path:
sys.path.insert(0, str(PACKAGE_PARENT))
from strategy29.backtest.window_analysis import slice_bundle
from strategy32.live.runtime import BEST_CASH_OVERLAY, LIVE_STRATEGY_OVERRIDES
from strategy32.research.soft_router import (
MacroScaleSpec,
build_cash_overlay_period_components,
compose_cash_overlay_curve,
load_component_bundle,
score_candidate,
segment_metrics,
)
CACHE_PATH = "/tmp/strategy32_fixed66_bundle.pkl"
OUT_JSON = Path("/tmp/strategy32_relaxed_macro_scaling_search.json")
RELAXED_OVERHEAT_OVERRIDES = {
**LIVE_STRATEGY_OVERRIDES,
"momentum_min_score": 0.58,
"momentum_min_relative_strength": -0.03,
"momentum_min_7d_return": 0.00,
"universe_min_avg_dollar_volume": 75_000_000.0,
"hard_filter_refresh_cadence": "1d",
"hard_filter_min_history_bars": 120,
"hard_filter_lookback_bars": 30,
"hard_filter_min_avg_dollar_volume": 50_000_000.0,
}
CURRENT_OVERHEAT_OVERRIDES = {
**LIVE_STRATEGY_OVERRIDES,
"hard_filter_refresh_cadence": "1d",
"hard_filter_min_history_bars": 120,
"hard_filter_lookback_bars": 30,
"hard_filter_min_avg_dollar_volume": 50_000_000.0,
}
WINDOWS = (
(365, "1y"),
(730, "2y"),
(1095, "3y"),
(1460, "4y"),
(1825, "5y"),
)
YEAR_PERIODS = (
("2021", pd.Timestamp("2021-03-16 04:00:00+00:00"), pd.Timestamp("2022-01-01 00:00:00+00:00")),
("2022", pd.Timestamp("2022-01-01 00:00:00+00:00"), pd.Timestamp("2023-01-01 00:00:00+00:00")),
("2023", pd.Timestamp("2023-01-01 00:00:00+00:00"), pd.Timestamp("2024-01-01 00:00:00+00:00")),
("2024", pd.Timestamp("2024-01-01 00:00:00+00:00"), pd.Timestamp("2025-01-01 00:00:00+00:00")),
("2025", pd.Timestamp("2025-01-01 00:00:00+00:00"), pd.Timestamp("2026-01-01 00:00:00+00:00")),
)
YTD_START = pd.Timestamp("2026-01-01 00:00:00+00:00")
def _clip01(value: float) -> float:
return min(max(float(value), 0.0), 1.0)
def _ramp(value: float, start: float, end: float) -> float:
if end == start:
return 1.0 if value >= end else 0.0
if value <= start:
return 0.0
if value >= end:
return 1.0
return (value - start) / (end - start)
def _build_macro_scale_map(sliced_bundle, *, timestamps: list[pd.Timestamp], spec: MacroScaleSpec) -> pd.Series:
btc_prices = sliced_bundle.prices["BTC"]
closes = btc_prices.set_index("timestamp")["close"].astype(float).sort_index()
daily = closes.resample("1D").last().dropna()
weekly = daily.resample("W-SUN").last().dropna()
fast = weekly.ewm(span=spec.fast_weeks, adjust=False).mean()
slow = weekly.ewm(span=spec.slow_weeks, adjust=False).mean()
close_scale = (weekly / slow - 1.0).apply(lambda value: _ramp(float(value), spec.close_gap_start, spec.close_gap_full))
fast_scale = (fast / slow - 1.0).apply(lambda value: _ramp(float(value), spec.fast_gap_start, spec.fast_gap_full))
blended = spec.close_weight * close_scale + (1.0 - spec.close_weight) * fast_scale
macro_scale = spec.floor + (1.0 - spec.floor) * blended.clip(0.0, 1.0)
aligned = macro_scale.reindex(pd.DatetimeIndex(timestamps, name="timestamp"), method="ffill")
return aligned.fillna(1.0).clip(spec.floor, 1.0).astype(float)
def _candidate_specs() -> list[MacroScaleSpec]:
specs: list[MacroScaleSpec] = []
for floor in (0.25, 0.35, 0.45):
for close_gap_start, close_gap_full in ((-0.08, 0.02), (-0.06, 0.02), (-0.05, 0.04)):
for fast_gap_start, fast_gap_full in ((-0.04, 0.01), (-0.03, 0.02)):
for close_weight in (0.55, 0.65):
specs.append(
MacroScaleSpec(
floor=floor,
close_gap_start=close_gap_start,
close_gap_full=close_gap_full,
fast_gap_start=fast_gap_start,
fast_gap_full=fast_gap_full,
close_weight=close_weight,
)
)
return specs
def _collect_metrics(curve: pd.Series, latest_bar: pd.Timestamp) -> tuple[dict[str, dict[str, float]], dict[str, dict[str, float]], float, int, int]:
window_results: dict[str, dict[str, float]] = {}
for days, label in WINDOWS:
start = latest_bar - pd.Timedelta(days=days)
window_results[label] = segment_metrics(curve, start, latest_bar)
year_results: dict[str, dict[str, float]] = {}
for label, start, end_exclusive in YEAR_PERIODS:
year_results[label] = segment_metrics(curve, start, min(latest_bar, end_exclusive - pd.Timedelta(seconds=1)))
year_results["2026_YTD"] = segment_metrics(curve, YTD_START, latest_bar)
score, negative_years, mdd_violations = score_candidate(
{label: window_results[label] for _, label in WINDOWS},
{label: year_results[label] for label, _, _ in YEAR_PERIODS},
)
return window_results, year_results, score, negative_years, mdd_violations
def _evaluate_exact_sequential(
bundle,
latest_bar: pd.Timestamp,
*,
core_overrides: dict[str, object],
macro_scale_spec: MacroScaleSpec | None,
) -> dict[str, object]:
window_results: dict[str, dict[str, float]] = {}
year_results: dict[str, dict[str, float]] = {}
periods = [
*(("window", label, latest_bar - pd.Timedelta(days=days), latest_bar) for days, label in WINDOWS),
*(("year", label, start, min(latest_bar, end_exclusive - pd.Timedelta(seconds=1))) for label, start, end_exclusive in YEAR_PERIODS),
("year", "2026_YTD", YTD_START, latest_bar),
]
latest_weights: list[dict[str, object]] = []
for kind, label, start, end in periods:
components = build_cash_overlay_period_components(
bundle=bundle,
eval_start=start,
eval_end=end,
profile_name=BEST_CASH_OVERLAY.regime_profile,
core_filter=BEST_CASH_OVERLAY.core_filter,
cap_engine=BEST_CASH_OVERLAY.cap_engine,
chop_engine=BEST_CASH_OVERLAY.chop_engine,
dist_engine=BEST_CASH_OVERLAY.dist_engine,
core_config_overrides=core_overrides,
macro_scale_spec=macro_scale_spec,
)
curve, weights = compose_cash_overlay_curve(candidate=BEST_CASH_OVERLAY, **components)
metrics = segment_metrics(curve, start, end)
if kind == "window":
window_results[label] = metrics
else:
year_results[label] = metrics
if label == "2026_YTD":
latest_weights = weights.tail(1).assign(timestamp=lambda df: df["timestamp"].astype(str)).to_dict(orient="records")
score, negative_years, mdd_violations = score_candidate(
{label: window_results[label] for _, label in WINDOWS},
{label: year_results[label] for label, _, _ in YEAR_PERIODS},
)
return {
"candidate": asdict(BEST_CASH_OVERLAY),
"core_overrides": core_overrides,
"macro_scale_spec": asdict(macro_scale_spec) if macro_scale_spec is not None else None,
"score": score,
"negative_years": negative_years,
"mdd_violations": mdd_violations,
"windows": window_results,
"years": year_results,
"latest_weights": latest_weights,
"validation": "exact_independent_periods_cash_overlay_sequential",
}
def main() -> None:
bundle, latest_bar = load_component_bundle(CACHE_PATH)
eval_start = latest_bar - pd.Timedelta(days=1825)
sliced = slice_bundle(bundle, eval_start - pd.Timedelta(days=365), latest_bar)
print("[phase] build relaxed core components", flush=True)
relaxed_components = build_cash_overlay_period_components(
bundle=bundle,
eval_start=eval_start,
eval_end=latest_bar,
profile_name=BEST_CASH_OVERLAY.regime_profile,
core_filter=BEST_CASH_OVERLAY.core_filter,
cap_engine=BEST_CASH_OVERLAY.cap_engine,
chop_engine=BEST_CASH_OVERLAY.chop_engine,
dist_engine=BEST_CASH_OVERLAY.dist_engine,
core_config_overrides=RELAXED_OVERHEAT_OVERRIDES,
)
print("[phase] search macro specs", flush=True)
search_rows: list[dict[str, object]] = []
specs = _candidate_specs()
for idx, spec in enumerate(specs, start=1):
macro_scale_map = _build_macro_scale_map(sliced, timestamps=relaxed_components["timestamps"][:-1], spec=spec)
curve, _weights = compose_cash_overlay_curve(
candidate=BEST_CASH_OVERLAY,
timestamps=relaxed_components["timestamps"],
score_frame=relaxed_components["score_frame"],
core_returns=relaxed_components["core_returns"],
core_exposure_frame=relaxed_components["core_exposure_frame"],
cap_returns=relaxed_components["cap_returns"],
chop_returns=relaxed_components["chop_returns"],
dist_returns=relaxed_components["dist_returns"],
macro_scale_map=macro_scale_map,
)
windows, years, score, negative_years, mdd_violations = _collect_metrics(curve, latest_bar)
search_rows.append(
{
"macro_scale_spec": asdict(spec),
"windows": windows,
"years": years,
"score": score,
"negative_years": negative_years,
"mdd_violations": mdd_violations,
}
)
if idx % 6 == 0 or idx == len(specs):
print(f"[search] {idx}/{len(specs)}", flush=True)
search_rows.sort(key=lambda row: float(row["score"]), reverse=True)
top_search = search_rows[:5]
search_only = os.getenv("STRATEGY32_SEARCH_ONLY", "").strip().lower() in {"1", "true", "yes", "on"}
if search_only:
payload = {
"analysis": "relaxed_overheat_macro_scaling_search",
"mode": "search_only",
"latest_bar": str(latest_bar),
"core_filter": "relaxed_overheat",
"candidate": asdict(BEST_CASH_OVERLAY),
"search_top": top_search,
}
OUT_JSON.write_text(json.dumps(payload, indent=2), encoding="utf-8")
print(json.dumps(payload, indent=2))
print(f"[saved] {OUT_JSON}")
return
print("[phase] exact baselines", flush=True)
baselines = {
"current_overheat": _evaluate_exact_sequential(
bundle,
latest_bar,
core_overrides=CURRENT_OVERHEAT_OVERRIDES,
macro_scale_spec=None,
),
"relaxed_overheat": _evaluate_exact_sequential(
bundle,
latest_bar,
core_overrides=RELAXED_OVERHEAT_OVERRIDES,
macro_scale_spec=None,
),
}
best_spec = MacroScaleSpec(**top_search[0]["macro_scale_spec"])
print(f"[phase] exact best spec {best_spec.name}", flush=True)
best_exact = _evaluate_exact_sequential(
bundle,
latest_bar,
core_overrides=RELAXED_OVERHEAT_OVERRIDES,
macro_scale_spec=best_spec,
)
payload = {
"analysis": "relaxed_overheat_macro_scaling_search",
"latest_bar": str(latest_bar),
"core_filter": "relaxed_overheat",
"candidate": asdict(BEST_CASH_OVERLAY),
"baselines": baselines,
"search_top": top_search,
"best_exact": best_exact,
}
OUT_JSON.write_text(json.dumps(payload, indent=2), encoding="utf-8")
print(json.dumps(payload, indent=2))
print(f"[saved] {OUT_JSON}")
if __name__ == "__main__":
main()