File size: 20,523 Bytes
2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 2c67d05 7db56ad 2c67d05 46590b0 2c67d05 46590b0 2c67d05 46590b0 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 |
"""
FinRL Agent for Algorithmic Trading
This module provides a FinRL-based reinforcement learning agent that can be integrated
with the existing algorithmic trading system. It supports various RL algorithms
including PPO, A2C, DDPG, and TD3, and can work with Alpaca broker for real trading.
"""
import numpy as np
import pandas as pd
import gymnasium as gym
from gymnasium import spaces
from stable_baselines3 import PPO, A2C, DDPG, TD3
from stable_baselines3.common.vec_env import DummyVecEnv
from stable_baselines3.common.callbacks import EvalCallback
import torch
import logging
from typing import Dict, List, Tuple, Optional, Any
from dataclasses import dataclass
import yaml
logger = logging.getLogger(__name__)
@dataclass
class FinRLConfig:
"""Configuration for FinRL agent"""
algorithm: str = "PPO" # PPO, A2C, DDPG, TD3
learning_rate: float = 0.0003
batch_size: int = 64
buffer_size: int = 1000000
learning_starts: int = 100
gamma: float = 0.99
tau: float = 0.005
train_freq: int = 1
gradient_steps: int = 1
target_update_interval: int = 1
exploration_fraction: float = 0.1
exploration_initial_eps: float = 1.0
exploration_final_eps: float = 0.05
max_grad_norm: float = 10.0
verbose: int = 1
tensorboard_log: str = "logs/finrl_tensorboard"
class TradingEnvironment(gym.Env):
"""
Custom trading environment for FinRL
This environment simulates a trading scenario where the agent can:
- Buy, sell, or hold positions
- Use technical indicators for decision making
- Manage portfolio value and risk
- Integrate with Alpaca broker for real trading
"""
def __init__(self, data: pd.DataFrame, config: Dict[str, Any],
initial_balance: float = 100000, transaction_fee: float = 0.001,
max_position: int = 100, use_real_broker: bool = False):
super().__init__()
self.data = data
self.config = config
self.initial_balance = initial_balance
self.transaction_fee = transaction_fee
self.max_position = max_position
self.use_real_broker = use_real_broker
# Initialize Alpaca broker if using real trading
self.alpaca_broker = None
if use_real_broker:
try:
from .alpaca_broker import AlpacaBroker
self.alpaca_broker = AlpacaBroker(config)
logger.info("Alpaca broker initialized for FinRL environment")
except Exception as e:
logger.error(f"Failed to initialize Alpaca broker: {e}")
self.use_real_broker = False
# Reset state
self.reset()
# Define action space: [-1, 0, 1] for sell, hold, buy
self.action_space = spaces.Discrete(3)
# Define observation space
# Features: OHLCV + technical indicators + portfolio state
n_features = len(self._get_features(self.data.iloc[0]))
self.observation_space = spaces.Box(
low=-np.inf, high=np.inf, shape=(n_features,), dtype=np.float32
)
def _get_features(self, row: pd.Series) -> np.ndarray:
"""Extract features from market data row"""
features = []
# Price features
features.extend([
row['open'], row['high'], row['low'], row['close'], row['volume']
])
# Technical indicators (if available)
for indicator in ['sma_20', 'sma_50', 'rsi', 'bb_upper', 'bb_lower', 'macd']:
if indicator in row.index:
features.append(row[indicator])
else:
features.append(0.0)
# Portfolio state
features.extend([
self.balance,
self.position,
self.portfolio_value,
self.total_return
])
return np.array(features, dtype=np.float32)
def _calculate_portfolio_value(self) -> float:
"""Calculate current portfolio value"""
current_price = self.data.iloc[self.current_step]['close']
return self.balance + (self.position * current_price)
def _calculate_reward(self) -> float:
"""Calculate reward based on portfolio performance"""
current_value = self._calculate_portfolio_value()
previous_value = self.previous_portfolio_value
# Calculate return
if previous_value > 0:
return (current_value - previous_value) / previous_value
else:
return 0.0
def step(self, action: int) -> Tuple[np.ndarray, float, bool, bool, Dict]:
"""Execute one step in the environment"""
# Get current market data
current_data = self.data.iloc[self.current_step]
current_price = current_data['close']
# Execute action
if action == 0: # Sell
if self.position > 0:
shares_to_sell = min(self.position, self.max_position)
if self.use_real_broker and self.alpaca_broker:
# Execute real order with Alpaca
result = self.alpaca_broker.place_market_order(
symbol=self.config['trading']['symbol'],
quantity=shares_to_sell,
side='sell'
)
if result['success']:
sell_value = result['filled_avg_price'] * shares_to_sell * (1 - self.transaction_fee)
self.balance += sell_value
self.position -= shares_to_sell
logger.info(f"Real sell order executed: {result['order_id']}")
else:
logger.warning(f"Real sell order failed: {result['error']}")
else:
# Simulate order execution
sell_value = shares_to_sell * current_price * (1 - self.transaction_fee)
self.balance += sell_value
self.position -= shares_to_sell
elif action == 2: # Buy
if self.balance > 0:
max_shares = min(
int(self.balance / current_price),
self.max_position - self.position
)
if max_shares > 0:
if self.use_real_broker and self.alpaca_broker:
# Execute real order with Alpaca
result = self.alpaca_broker.place_market_order(
symbol=self.config['trading']['symbol'],
quantity=max_shares,
side='buy'
)
if result['success']:
buy_value = result['filled_avg_price'] * max_shares * (1 + self.transaction_fee)
self.balance -= buy_value
self.position += max_shares
logger.info(f"Real buy order executed: {result['order_id']}")
else:
logger.warning(f"Real buy order failed: {result['error']}")
else:
# Simulate order execution
buy_value = max_shares * current_price * (1 + self.transaction_fee)
self.balance -= buy_value
self.position += max_shares
# Update portfolio value
self.previous_portfolio_value = self.portfolio_value
self.portfolio_value = self._calculate_portfolio_value()
self.total_return = (self.portfolio_value - self.initial_balance) / self.initial_balance
# Move to next step
self.current_step += 1
# Check if episode is done
done = self.current_step >= len(self.data) - 1
# Get observation for next step
if not done:
observation = self._get_features(self.data.iloc[self.current_step])
else:
observation = self._get_features(self.data.iloc[-1])
# Calculate reward
reward = self._calculate_reward()
# Additional info
info = {
'portfolio_value': self.portfolio_value,
'total_return': self.total_return,
'position': self.position,
'balance': self.balance,
'step': self.current_step
}
return observation, reward, done, False, info
def reset(self, seed: Optional[int] = None) -> Tuple[np.ndarray, Dict]:
"""Reset the environment"""
super().reset(seed=seed)
self.current_step = 0
self.balance = self.initial_balance
self.position = 0
self.portfolio_value = self.initial_balance
self.previous_portfolio_value = self.initial_balance
self.total_return = 0.0
# Get initial observation
observation = self._get_features(self.data.iloc[0])
return observation, {}
class FinRLAgent:
"""
FinRL-based reinforcement learning agent for algorithmic trading
"""
def __init__(self, config: FinRLConfig):
self.config = config
self.model = None
self.env = None
self.eval_env = None
self.callback = None
logger.info(f"Initializing FinRL agent with algorithm: {config.algorithm}")
def create_environment(self, data: pd.DataFrame, config: Dict[str, Any],
initial_balance: float = 100000, use_real_broker: bool = False) -> TradingEnvironment:
"""Create trading environment from market data"""
return TradingEnvironment(
data=data,
config=config,
initial_balance=initial_balance,
transaction_fee=0.001,
max_position=100,
use_real_broker=use_real_broker
)
def prepare_data(self, data: pd.DataFrame) -> pd.DataFrame:
"""Prepare data with technical indicators for FinRL"""
df = data.copy()
# Add technical indicators if not present
if 'sma_20' not in df.columns:
df['sma_20'] = df['close'].rolling(window=20).mean()
if 'sma_50' not in df.columns:
df['sma_50'] = df['close'].rolling(window=50).mean()
if 'rsi' not in df.columns:
df['rsi'] = self._calculate_rsi(df['close'])
if 'bb_upper' not in df.columns or 'bb_lower' not in df.columns:
bb_upper, bb_lower = self._calculate_bollinger_bands(df['close'])
df['bb_upper'] = bb_upper
df['bb_lower'] = bb_lower
if 'macd' not in df.columns:
df['macd'] = self._calculate_macd(df['close'])
# Fill NaN values
df = df.bfill().fillna(0)
return df
def train(self, data: pd.DataFrame, config: Dict[str, Any],
total_timesteps: int = 100000, use_real_broker: bool = False) -> Dict[str, Any]:
"""
Train the FinRL agent
Args:
data: Market data for training
config: Configuration dictionary
total_timesteps: Number of timesteps for training
use_real_broker: Whether to use real Alpaca broker during training
Returns:
Training results dictionary
"""
try:
# Prepare data
prepared_data = self.prepare_data(data)
# Create environment
self.env = self.create_environment(prepared_data, config, use_real_broker=use_real_broker)
# Create evaluation environment (without real broker)
eval_data = prepared_data.copy()
self.eval_env = self.create_environment(eval_data, config, use_real_broker=False)
# Create callback for evaluation
self.callback = EvalCallback(
self.eval_env,
best_model_save_path=config['finrl']['training']['model_save_path'],
log_path=config['finrl']['tensorboard_log'],
eval_freq=config['finrl']['training']['eval_freq'],
deterministic=True,
render=False
)
# Initialize model based on algorithm
if self.config.algorithm == "PPO":
self.model = PPO(
"MlpPolicy",
self.env,
learning_rate=self.config.learning_rate,
batch_size=self.config.batch_size,
buffer_size=self.config.buffer_size,
learning_starts=self.config.learning_starts,
gamma=self.config.gamma,
train_freq=self.config.train_freq,
gradient_steps=self.config.gradient_steps,
verbose=self.config.verbose,
tensorboard_log=self.config.tensorboard_log
)
elif self.config.algorithm == "A2C":
self.model = A2C(
"MlpPolicy",
self.env,
learning_rate=self.config.learning_rate,
gamma=self.config.gamma,
verbose=self.config.verbose,
tensorboard_log=self.config.tensorboard_log
)
elif self.config.algorithm == "DDPG":
self.model = DDPG(
"MlpPolicy",
self.env,
learning_rate=self.config.learning_rate,
buffer_size=self.config.buffer_size,
learning_starts=self.config.learning_starts,
gamma=self.config.gamma,
tau=self.config.tau,
train_freq=self.config.train_freq,
gradient_steps=self.config.gradient_steps,
verbose=self.config.verbose,
tensorboard_log=self.config.tensorboard_log
)
elif self.config.algorithm == "TD3":
self.model = TD3(
"MlpPolicy",
self.env,
learning_rate=self.config.learning_rate,
buffer_size=self.config.buffer_size,
learning_starts=self.config.learning_starts,
gamma=self.config.gamma,
tau=self.config.tau,
train_freq=self.config.train_freq,
gradient_steps=self.config.gradient_steps,
verbose=self.config.verbose,
tensorboard_log=self.config.tensorboard_log
)
else:
raise ValueError(f"Unsupported algorithm: {self.config.algorithm}")
# Train the model
logger.info(f"Starting training with {total_timesteps} timesteps")
self.model.learn(
total_timesteps=total_timesteps,
callback=self.callback,
progress_bar=True
)
# Save the final model
model_path = f"{config['finrl']['training']['model_save_path']}/final_model"
self.model.save(model_path)
logger.info(f"Training completed. Model saved to {model_path}")
return {
'success': True,
'algorithm': self.config.algorithm,
'total_timesteps': total_timesteps,
'model_path': model_path
}
except Exception as e:
logger.error(f"Error during training: {e}")
return {
'success': False,
'error': str(e)
}
def predict(self, data: pd.DataFrame, config: Dict[str, Any],
use_real_broker: bool = False) -> Dict[str, Any]:
"""
Make predictions using the trained model
Args:
data: Market data for prediction
config: Configuration dictionary
use_real_broker: Whether to use real Alpaca broker for execution
Returns:
Prediction results dictionary
"""
try:
if self.model is None:
# Try to load model
model_path = config['finrl']['inference']['model_path']
if config['finrl']['inference']['use_trained_model']:
self.model = self._load_model(model_path, config)
if self.model is None:
return {'success': False, 'error': 'No trained model available'}
else:
return {'success': False, 'error': 'No model available for prediction'}
# Prepare data
prepared_data = self.prepare_data(data)
# Create environment
env = self.create_environment(prepared_data, config, use_real_broker=use_real_broker)
# Run prediction
obs, _ = env.reset()
done = False
actions = []
rewards = []
portfolio_values = []
while not done:
action, _ = self.model.predict(obs, deterministic=True)
obs, reward, done, _, info = env.step(action)
actions.append(action)
rewards.append(reward)
portfolio_values.append(info['portfolio_value'])
# Calculate final metrics
initial_value = config['trading']['capital']
final_value = portfolio_values[-1] if portfolio_values else initial_value
total_return = (final_value - initial_value) / initial_value
return {
'success': True,
'actions': actions,
'rewards': rewards,
'portfolio_values': portfolio_values,
'initial_value': initial_value,
'final_value': final_value,
'total_return': total_return,
'total_trades': len([a for a in actions if a != 1]) # Count non-hold actions
}
except Exception as e:
logger.error(f"Error during prediction: {e}")
return {
'success': False,
'error': str(e)
}
def _load_model(self, model_path: str, config: Dict[str, Any]):
"""Load a trained model"""
try:
if config['finrl']['algorithm'] == "PPO":
return PPO.load(model_path)
elif config['finrl']['algorithm'] == "A2C":
return A2C.load(model_path)
elif config['finrl']['algorithm'] == "DDPG":
return DDPG.load(model_path)
elif config['finrl']['algorithm'] == "TD3":
return TD3.load(model_path)
else:
logger.error(f"Unsupported algorithm for model loading: {config['finrl']['algorithm']}")
return None
except Exception as e:
logger.error(f"Error loading model: {e}")
return None
def _calculate_rsi(self, prices: pd.Series, period: int = 14) -> pd.Series:
"""Calculate RSI indicator"""
delta = prices.diff()
gain = (delta.where(delta > 0, 0)).rolling(window=period).mean()
loss = (-delta.where(delta < 0, 0)).rolling(window=period).mean()
rs = gain / loss
return 100 - (100 / (1 + rs))
def _calculate_bollinger_bands(self, prices: pd.Series, period: int = 20, std_dev: int = 2) -> Tuple[pd.Series, pd.Series]:
"""Calculate Bollinger Bands"""
sma = prices.rolling(window=period).mean()
std = prices.rolling(window=period).std()
upper_band = sma + (std * std_dev)
lower_band = sma - (std * std_dev)
return upper_band, lower_band
def _calculate_macd(self, prices: pd.Series, fast: int = 12, slow: int = 26, signal: int = 9) -> pd.Series:
"""Calculate MACD indicator"""
ema_fast = prices.ewm(span=fast).mean()
ema_slow = prices.ewm(span=slow).mean()
macd = ema_fast - ema_slow
return macd |