Files

138 lines
7.8 KiB
Python

"""Connect raw daily bars to market context, PIT breadth, features and RS."""
from __future__ import annotations
import numpy as np
import pandas as pd
from .config import RelativeStrengthWeights, ResearchSettings
from .indicators import (breadth_series, equal_weight_index, market_regimes, mcclellan,
price_features, relative_strength)
from .setups import candidate_features
from .universe import PointInTimeUniverse
def _equal_weight_bars(bars_by_symbol: dict[str, pd.DataFrame], closes: pd.DataFrame,
members_by_date: dict[pd.Timestamp, list[str]], base: float = 100.0) -> pd.DataFrame:
columns = {field: [] for field in ("open", "high", "low", "close", "volume")}
level = base
previous_close = closes.shift(1)
for date in closes.index:
members = [symbol for symbol in members_by_date.get(pd.Timestamp(date), []) if symbol in bars_by_symbol]
valid = [s for s in members if date in bars_by_symbol[s].index and date in previous_close.index
and pd.notna(previous_close.loc[date, s]) and previous_close.loc[date, s] > 0]
if not valid:
daily_open = daily_high = daily_low = daily_close = 0.0
volume = 0.0
else:
fields = {field: [] for field in ("open", "high", "low", "close")}
volume = 0.0
for symbol in valid:
bar = bars_by_symbol[symbol].loc[date]
previous = float(previous_close.loc[date, symbol])
for field in fields:
fields[field].append(float(bar[field]) / previous - 1)
volume += float(bar.get("volume", 0.0))
daily_open, daily_high, daily_low, daily_close = (float(np.nanmean(fields[field])) for field in ("open", "high", "low", "close"))
daily_high = max(daily_high, daily_open, daily_close)
daily_low = min(daily_low, daily_open, daily_close)
if columns["close"]:
opening, high, low, close = (level * (1 + value) for value in (daily_open, daily_high, daily_low, daily_close))
level = close
else:
opening = high = low = close = level
columns["open"].append(opening)
columns["high"].append(max(high, opening, close))
columns["low"].append(min(low, opening, close))
columns["close"].append(close)
columns["volume"].append(volume)
return pd.DataFrame(columns, index=closes.index)
def build_market_context(bars_by_symbol: dict[str, pd.DataFrame], set_bars: pd.DataFrame,
membership: PointInTimeUniverse | None = None,
current_members: list[str] | None = None,
*, accurate: bool = False,
research: ResearchSettings | None = None) -> tuple[pd.DataFrame, pd.Series, pd.DataFrame]:
"""Build SET, equal-weight proxy, breadth/MCO/MCSI and deterministic regime."""
research = research or ResearchSettings()
if not bars_by_symbol:
raise ValueError("at least one constituent price series is required")
dates = set_bars.index
stock_closes = pd.DataFrame({symbol: frame["close"].reindex(dates) for symbol, frame in bars_by_symbol.items()}, index=dates)
if accurate:
if membership is None:
raise ValueError("accurate mode requires a point-in-time membership file")
members_by_date = membership.members_by_date(dates)
elif membership is not None:
anchor = dates[-1]
members = membership.members(anchor)
if not members:
members = membership.members(max(row.end for row in membership.periods))
members_by_date = {pd.Timestamp(date): members for date in dates}
else:
if not current_members:
raise ValueError("fast mode requires a supplied constituent snapshot")
members = [symbol.upper().removesuffix(".BK") for symbol in current_members]
members_by_date = {pd.Timestamp(date): members for date in dates}
closes = stock_closes.copy()
breadth = breadth_series(closes, members_by_date)
ew_bars = _equal_weight_bars(bars_by_symbol, closes, members_by_date)
ew_features = price_features(ew_bars, slope_lookback=research.slope_lookback,
atr_period=research.atr_period, adr_period=research.adr_period)
ew = ew_features["close"].rename("set100ew_close")
set_features = price_features(set_bars, slope_lookback=research.slope_lookback,
atr_period=research.atr_period, adr_period=research.adr_period)
mco_mcsi = mcclellan(breadth, short_span=research.mco_short_span,
long_span=research.mco_long_span, z_window=research.mco_z_window)
context = set_features.join(breadth).join(mco_mcsi)
context["set100ew_close"] = ew
context["set100ew_ema21"] = ew_features["ema21_close"]
context["set100ew_trend_state"] = ew_features["trend_state"]
context["ew_trend_state"] = ew_features["trend_state"]
context = market_regimes(context, overbought_extension_atr=research.overbought_extension_atr,
repair_sigma=research.mco_repair_sigma)
context.attrs["members_by_date"] = members_by_date
return context, ew, breadth
def build_feature_panel(bars_by_symbol: dict[str, pd.DataFrame], set_bars: pd.DataFrame,
market: pd.DataFrame, *, rs_weights: RelativeStrengthWeights | None = None,
volume_ratio: float = 1.0,
research: ResearchSettings | None = None) -> dict[str, pd.DataFrame]:
research = research or ResearchSettings()
weights = rs_weights or RelativeStrengthWeights()
close_series = {symbol: frame["close"] for symbol, frame in bars_by_symbol.items()}
set100ew = market["set100ew_close"]
set_close = market["close"]
rs = relative_strength(close_series, set100ew,
weights={"1m": weights.one_month, "3m": weights.three_month,
"12m": weights.twelve_month, "52w": weights.high_52w},
set_benchmark=set_close,
high_prices={symbol: frame["high"] for symbol, frame in bars_by_symbol.items()},
members_by_date=market.attrs.get("members_by_date"))
market_columns = ["regime", "regime_reasons", "new_risk_allowed", "net_advances", "advancing",
"declining", "unchanged", "mco", "mco_mean", "mco_std", "mco_z", "mcsi",
"mcsi_10dma", "mcsi_state", "set100ew_close", "set100ew_ema21", "ew_trend_state"]
context_fields = market[[column for column in market_columns if column in market.columns]]
output: dict[str, pd.DataFrame] = {}
for symbol, bars in bars_by_symbol.items():
features = candidate_features(
price_features(bars, slope_lookback=research.slope_lookback,
atr_period=research.atr_period, adr_period=research.adr_period),
pullback_window=research.pullback_volume_window,
tightness_window=research.tightness_window,
volume_ratio=volume_ratio, pivot_left=research.pivot_left_bars,
pivot_right=research.pivot_right_bars)
features["symbol"] = symbol
rs_columns = [f"{symbol}_{suffix}" for suffix in (
"rs_percentile", "rs_1m", "rs_3m", "rs_12m", "rs_52w",
"rs_1m_vs_set", "rs_3m_vs_set", "rs_12m_vs_set",
"rs_1m_vs_set100ew", "rs_3m_vs_set100ew", "rs_12m_vs_set100ew", "52w_proximity_pct")]
rs_part = rs[[column for column in rs_columns if column in rs.columns]].rename(
columns={column: column.removeprefix(f"{symbol}_") for column in rs_columns if column in rs.columns})
features = features.join(rs_part.reindex(features.index))
features = features.join(context_fields.reindex(features.index))
output[symbol] = features
return output