File size: 8,863 Bytes
fc115d5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 | """
Multi-Timeframe Trading Environment
Extends AdvancedTradingEnv with real multi-timeframe observations:
- 15-minute data: micro order flow, entry timing, short-term momentum
- 1-hour data: main trading timeframe (existing)
- 4-hour data: macro bias, trend direction, support/resistance
The observation is extended with compact MTF summary features
(not full bars), keeping the observation space manageable.
"""
import gymnasium as gym
from gymnasium import spaces
import numpy as np
import pandas as pd
from typing import Optional, Dict, Tuple, List
import logging
from .advanced_env import AdvancedTradingEnv, Actions, Positions
from .advanced_features import AdvancedFeatureEngine
from .advanced_rewards import AdvancedRewardCalculator
logger = logging.getLogger(__name__)
def resample_to_higher_tf(df_1h: pd.DataFrame, target_tf: str = '4h') -> pd.DataFrame:
"""
Resample 1h OHLCV data to a higher timeframe.
Args:
df_1h: DataFrame with 1h OHLCV data (DatetimeIndex)
target_tf: Target timeframe ('4h', '1d', etc.)
Returns:
Resampled DataFrame
"""
# Map our labels to pandas resample rules
tf_map = {'4h': '4h', '1d': '1D', '1w': '1W'}
rule = tf_map.get(target_tf, target_tf)
resampled = df_1h.resample(rule).agg({
'open': 'first',
'high': 'max',
'low': 'min',
'close': 'last',
'volume': 'sum',
}).dropna()
return resampled
def compute_mtf_features_at(
df_4h: pd.DataFrame,
timestamp: pd.Timestamp,
lookback: int = 30,
) -> np.ndarray:
"""
Compute compact 4h timeframe features at a given 1h timestamp.
Returns 12 features:
- 4h trend (EMA 7 vs 14 slope)
- 4h RSI
- 4h momentum (return over 5, 10, 20 bars)
- 4h volatility (ATR ratio)
- 4h volume trend
- 4h MACD state
- 4h Bollinger position
- 4h support/resistance distance
- 4h candle body ratio
- 4h trend strength (ADX proxy)
"""
# Find the most recent 4h bar at or before this timestamp
mask = df_4h.index <= timestamp
available = df_4h[mask]
if len(available) < lookback:
return np.zeros(12, dtype=np.float32)
window = available.tail(lookback)
close = window['close'].values
high = window['high'].values
low = window['low'].values
volume = window['volume'].values
opn = window['open'].values
features = np.zeros(12, dtype=np.float32)
try:
# 1. EMA trend: EMA7 vs EMA14 normalized
ema7 = pd.Series(close).ewm(span=7).mean().values
ema14 = pd.Series(close).ewm(span=14).mean().values
features[0] = np.clip((ema7[-1] - ema14[-1]) / (close[-1] * 0.01 + 1e-10), -3, 3)
# 2. RSI 14
delta = np.diff(close)
gain = np.where(delta > 0, delta, 0)
loss = np.where(delta < 0, -delta, 0)
avg_gain = pd.Series(gain).rolling(14).mean().values[-1]
avg_loss = pd.Series(loss).rolling(14).mean().values[-1]
if avg_loss > 0:
rs = avg_gain / avg_loss
features[1] = (100 - 100 / (1 + rs)) / 100 # Normalized 0-1
else:
features[1] = 0.5
# 3-5. Momentum (returns over 5, 10, 20 bars)
for i, period in enumerate([5, 10, 20]):
if len(close) > period:
features[2 + i] = np.clip((close[-1] / close[-period] - 1) * 10, -3, 3)
# 6. ATR ratio (current vs average)
tr = np.maximum(high[1:] - low[1:],
np.maximum(np.abs(high[1:] - close[:-1]),
np.abs(low[1:] - close[:-1])))
atr_14 = pd.Series(tr).rolling(14).mean().values[-1]
atr_avg = pd.Series(tr).rolling(min(len(tr), 50)).mean().values[-1]
features[5] = np.clip(atr_14 / (atr_avg + 1e-10) - 1, -2, 2)
# 7. Volume trend (current vs 20-bar average)
vol_avg = np.mean(volume[-20:]) if len(volume) >= 20 else np.mean(volume)
features[6] = np.clip(volume[-1] / (vol_avg + 1e-10) - 1, -3, 3)
# 8. MACD state (sign of MACD histogram)
ema12 = pd.Series(close).ewm(span=12).mean().values
ema26 = pd.Series(close).ewm(span=26).mean().values
macd = ema12 - ema26
signal = pd.Series(macd).ewm(span=9).mean().values
hist = macd[-1] - signal[-1]
features[7] = np.clip(hist / (close[-1] * 0.001 + 1e-10), -3, 3)
# 9. Bollinger position
bb_ma = np.mean(close[-20:])
bb_std = np.std(close[-20:])
if bb_std > 0:
features[8] = np.clip((close[-1] - bb_ma) / (2 * bb_std + 1e-10), -1.5, 1.5)
# 10. Support/resistance distance (distance to recent high/low)
recent_high = np.max(high[-20:])
recent_low = np.min(low[-20:])
price_range = recent_high - recent_low
if price_range > 0:
features[9] = (close[-1] - recent_low) / price_range # 0-1 position
# 11. Candle body ratio (last bar)
hl_range = high[-1] - low[-1]
if hl_range > 0:
features[10] = (close[-1] - opn[-1]) / hl_range # -1 to 1
# 12. Trend strength (directional movement proxy)
if len(close) >= 14:
up_moves = np.diff(high[-15:])
down_moves = -np.diff(low[-15:])
up_moves = np.where(up_moves > 0, up_moves, 0)
down_moves = np.where(down_moves > 0, down_moves, 0)
di_plus = np.mean(up_moves[-14:])
di_minus = np.mean(down_moves[-14:])
if (di_plus + di_minus) > 0:
features[11] = np.clip((di_plus - di_minus) / (di_plus + di_minus), -1, 1)
except Exception as e:
logger.debug(f"MTF feature error: {e}")
return np.nan_to_num(features, nan=0.0, posinf=1.0, neginf=-1.0).astype(np.float32)
N_MTF_FEATURES = 12 # per higher timeframe
N_MTF_TOTAL = N_MTF_FEATURES # just 4h for now (15m would need separate data)
class MultiTimeframeTradingEnv(AdvancedTradingEnv):
"""
Multi-Timeframe Trading Environment.
Extends AdvancedTradingEnv by appending compact 4h timeframe features
to the observation space. The agent sees:
[flattened 1h OHLCV+features | agent state | 4h MTF features]
This is backward-compatible — the original observation is preserved
and the MTF features are appended at the end.
"""
def __init__(
self,
df: pd.DataFrame,
df_4h: pd.DataFrame = None,
initial_balance: float = 10000.0,
lookback_window: int = 48,
trading_fee: float = 0.0004,
position_size: float = 0.25,
max_position: int = 1,
render_mode: Optional[str] = None,
):
# Store 4h data before parent init
if df_4h is not None:
self._df_4h = df_4h.copy()
else:
# Auto-resample from 1h if not provided
self._df_4h = resample_to_higher_tf(df, '4h')
# Parent init (computes features, sets observation space)
super().__init__(
df=df,
initial_balance=initial_balance,
lookback_window=lookback_window,
trading_fee=trading_fee,
position_size=position_size,
max_position=max_position,
render_mode=render_mode,
)
# Extend observation space with MTF features
base_obs_dim = self.observation_space.shape[0]
new_obs_dim = base_obs_dim + N_MTF_TOTAL
self.observation_space = spaces.Box(
low=-np.inf, high=np.inf, shape=(new_obs_dim,), dtype=np.float32
)
logger.info(
f"🔀 MTF env initialized: base={base_obs_dim}, "
f"mtf={N_MTF_TOTAL}, total={new_obs_dim}, "
f"4h bars={len(self._df_4h)}"
)
def _get_observation(self) -> np.ndarray:
"""Construct observation with base features + MTF features."""
# Get base observation from parent
base_obs = super()._get_observation()
# Compute 4h MTF features at current timestamp
current_idx = min(self.current_step, len(self.df) - 1)
current_ts = self.df.index[current_idx]
mtf_4h = compute_mtf_features_at(self._df_4h, current_ts)
# Concatenate
observation = np.concatenate([base_obs, mtf_4h]).astype(np.float32)
return observation
def create_mtf_env(
df_1h: pd.DataFrame,
df_4h: pd.DataFrame = None,
config: Optional[Dict] = None,
) -> MultiTimeframeTradingEnv:
"""Factory function to create MTF environment."""
config = config or {}
return MultiTimeframeTradingEnv(
df=df_1h,
df_4h=df_4h,
initial_balance=config.get('initial_balance', 10000.0),
lookback_window=config.get('lookback_window', 48),
trading_fee=config.get('trading_fee', 0.0004),
position_size=config.get('position_size', 0.25),
)
|