#machinemodel.py import aiohttp import numpy as np import pandas as pd from…
- User
#machine_model.py import aiohttp import numpy as np import pandas as pd from sklearn.feature_selection import SelectKBest, mutual_info_regression from sklearn.metrics import mean_absolute_error from sklearn.model_selection import RandomizedSearchCV, train_test_split, TimeSeriesSplit from sklearn.preprocessing import StandardScaler from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor from sklearn.svm import SVR from xgboost import XGBRegressor from lightgbm import LGBMRegressor import tensorflow as tf from keras import Sequential from tensorflow.python.keras.layers.recurrent import LSTM from tensorflow.python.keras.layers import Dense, Dropout, Attention, Conv1D from keras.src.callbacks import EarlyStopping from tensorflow.python.keras.optimizer_v1 import Adam from statsmodels.tsa.arima.model import ARIMA from prophet import Prophet import talib import asyncio from typing import Dict, List, Tuple import logging from .fetch_data import DataFetcher # Assume this is appropriately implemented to fetch data from .models import Signal # Assume this is your Django model for saving signals import json
Logger configuration
logging.basicConfig(level=logging.INFO) logger = logging.getLogger(name)
class MachineTradingBot: def init(self, symbols: List[str]): self.models = { 'random_forest': RandomForestRegressor(random_state=42), 'svm': SVR(), 'xgboost': XGBRegressor(random_state=42), 'lightgbm': LGBMRegressor(random_state=42), 'gradient_boosting': GradientBoostingRegressor(random_state=42), 'lstm': None, 'hybrid_cnn_lstm': None, 'arima': None, 'prophet': Prophet(), } self.scaler = StandardScaler() self.best_params = {} self.feature_importances = {} self.model_weights = {model: 1 / len(self.models) for model in self.models} self.model_performance = {model: [] for model in self.models} self.timeframes = ['5m', '15m', '1h'] self.symbols = symbols self.session = None self.fetcher = DataFetcher(symbols=symbols, timeframes=self.timeframes)
async def initialize(self): self.session = aiohttp.ClientSession() await self.fetcher.initialize_exchange() async def fetch_data(self, symbol: str, timeframe: str) -> pd.DataFrame: # Fetch historical data using DataFetcher await self.fetcher.fetch_historical_data() # Placeholder for fetching real-time updates def preprocess_data(self, data: pd.DataFrame) -> pd.DataFrame: data = self.add_technical_indicators(data) data = self.add_fourier_features(data) data = self.add_lagged_features(data) data = self.add_market_regime_features(data) data.dropna(inplace=True) return data def add_technical_indicators(self, data: pd.DataFrame) -> pd.DataFrame: for col in ['open', 'high', 'low', 'close', 'volume']: if col not in data.columns: logger.warning(f"Column {col} not found in data. Skipping technical indicators.") return data try: data['rsi'] = talib.RSI(data['close']) data['macd'], data['macd_signal'], _ = talib.MACD(data['close']) data['bbands_upper'], data['bbands_middle'], data['bbands_lower'] = talib.BBANDS(data['close']) data['adx'] = talib.ADX(data['high'], data['low'], data['close']) data['cci'] = talib.CCI(data['high'], data['low'], data['close']) data['mfi'] = talib.MFI(data['high'], data['low'], data['close'], data['volume']) except Exception as e: logger.error(f"Error calculating technical indicators: {str(e)}") return data def add_fourier_features(self, data: pd.DataFrame) -> pd.DataFrame: for period in [5, 10, 22]: for order in range(1, 3): data[f'fourier_sin_{period}_{order}'] = np.sin(2 * np.pi * order * data.index.dayofyear / period) data[f'fourier_cos_{period}_{order}'] = np.cos(2 * np.pi * order * data.index.dayofyear / period) return data def add_lagged_features(self, data: pd.DataFrame) -> pd.DataFrame: for lag in [1, 3, 5, 7, 14, 30]: data[f'close_lag_{lag}'] = data['close'].shift(lag) data[f'volume_lag_{lag}'] = data['volume'].shift(lag) return data def add_market_regime_features(self, data: pd.DataFrame) -> pd.DataFrame: returns = data['close'].pct_change() volatility = returns.rolling(window=22).std() data['market_regime'] = pd.cut(volatility, bins=3, labels=['low', 'medium', 'high']) data = pd.get_dummies(data, columns=['market_regime']) return data def prepare_features(self, data: pd.DataFrame) -> Tuple[pd.DataFrame, pd.Series]: features = data.drop(['open', 'high', 'low', 'close', 'volume'], axis=1) target = data['close'].shift(-1) selector = SelectKBest(mutual_info_regression, k=50) features = pd.DataFrame(selector.fit_transform(features, target), columns=features.columns[selector.get_support()]) return features, target async def train_models(self, data: pd.DataFrame): features, target = self.prepare_features(data) X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, shuffle=False) X_train_scaled = self.scaler.fit_transform(X_train) X_test_scaled = self.scaler.transform(X_test) tasks = [] for model_name in ['random_forest', 'svm', 'xgboost', 'lightgbm', 'gradient_boosting']: tasks.append(self.tune_and_train(model_name, X_train_scaled, y_train, X_test_scaled, y_test)) tasks.append(self.train_lstm(X_train_scaled, y_train, X_test_scaled, y_test)) tasks.append(self.train_hybrid_cnn_lstm(X_train_scaled, y_train, X_test_scaled, y_test)) tasks.append(self.train_arima(data['close'])) prophet_data = pd.DataFrame({'ds': data.index, 'y': data['close']}) tasks.append(self.train_prophet(prophet_data)) await asyncio.gather(*tasks) self.calculate_feature_importances(features.columns) async def tune_and_train(self, model_name: str, X_train: np.ndarray, y_train: pd.Series, X_test: np.ndarray, y_test: pd.Series): logger.info(f"Tuning and training {model_name}...") param_distributions = self.get_param_distributions(model_name) tscv = TimeSeriesSplit(n_splits=5) search = RandomizedSearchCV( self.models[model_name], param_distributions=param_distributions, n_iter=50, cv=tscv, random_state=42, n_jobs=-1, scoring='neg_mean_absolute_error' ) search.fit(X_train, y_train) self.models[model_name] = search.best_estimator_ self.best_params[model_name] = search.best_params_ y_pred = self.models[model_name].predict(X_test) mae = mean_absolute_error(y_test, y_pred) self.model_performance[model_name].append(mae) logger.info(f"Best parameters for {model_name}: {search.best_params_}") logger.info(f"MAE for {model_name}: {mae}") async def train_lstm(self, X_train: np.ndarray, y_train: pd.Series, X_test: np.ndarray, y_test: pd.Series): model = Sequential([ LSTM(100, activation='relu', input_shape=(X_train.shape[1], 1), return_sequences=True), Attention(), Dropout(0.2), LSTM(50, activation='relu'), Dropout(0.2), Dense(1) ]) model.compile(optimizer=Adam(learning_rate=0.001), loss='mse') X_train_reshaped = X_train.reshape((X_train.shape[0], X_train.shape[1], 1)) X_test_reshaped = X_test.reshape((X_test.shape[0], X_test.shape[1], 1)) early_stopping = EarlyStopping(patience=10, restore_best_weights=True) model.fit( X_train_reshaped, y_train, epochs=100, batch_size=32, validation_split=0.2, callbacks=[early_stopping], verbose=0 ) y_pred = model.predict(X_test_reshaped) mae = mean_absolute_error(y_test, y_pred) self.model_performance['lstm'].append(mae) logger.info(f"MAE for LSTM: {mae}") self.models['lstm'] = model async def train_hybrid_cnn_lstm(self, X_train: np.ndarray, y_train: pd.Series, X_test: np.ndarray, y_test: pd.Series): model = Sequential([ Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(X_train.shape[1], 1)), Conv1D(filters=64, kernel_size=3, activation='relu'), LSTM(50, activation='relu', return_sequences=True), Attention(), LSTM(50, activation='relu'), Dense(1) ]) model.compile(optimizer=Adam(learning_rate=0.001), loss='mse') X_train_reshaped = X_train.reshape((X_train.shape[0], X_train.shape[1], 1)) X_test_reshaped = X_test.reshape((X_test.shape[0], X_test.shape[1], 1)) early_stopping = EarlyStopping(patience=10, restore_best_weights=True) model.fit( X_train_reshaped, y_train, epochs=100, batch_size=32, validation_split=0.2, callbacks=[early_stopping], verbose=0 ) y_pred = model.predict(X_test_reshaped) mae = mean_absolute_error(y_test, y_pred) self.model_performance['hybrid_cnn_lstm'].append(mae) logger.info(f"MAE for Hybrid CNN-LSTM: {mae}") self.models['hybrid_cnn_lstm'] = model async def train_arima(self, time_series: pd.Series): model = ARIMA(time_series, order=(5, 1, 0)) fitted = model.fit() # Evaluate performance y_pred = fitted.forecast(steps=len(time_series)) mae = mean_absolute_error(time_series, y_pred) self.model_performance['arima'].append(mae) logger.info(f"MAE for ARIMA: {mae}") self.models['arima'] = fitted async def train_prophet(self, prophet_data: pd.DataFrame): model = Prophet() model.fit(prophet_data) future = model.make_future_dataframe(periods=len(prophet_data)) forecast = model.predict(future) y_pred = forecast['yhat'][-len(prophet_data):] mae = mean_absolute_error(prophet_data['y'], y_pred) self.model_performance['prophet'].append(mae) logger.info(f"MAE for Prophet: {mae}") self.models['prophet'] = model async def generate_signals(self, latest_data: pd.DataFrame) -> Tuple[Dict[str, str], Dict[str, float]]: features, _ = self.prepare_features(latest_data) scaled_features = self.scaler.transform(features) signals = {} confidences = {} for model_name in ['random_forest', 'svm', 'xgboost', 'lightgbm', 'gradient_boosting']: prediction = self.models[model_name].predict(scaled_features) signal_strength = prediction[-1] - latest_data['close'].iloc[-1] category = "strong buy" if signal_strength > 0.005 else "weak buy" if signal_strength > 0 else "strong sell" if signal_strength < -0.005 else "weak sell" signals[model_name] = category confidences[model_name] = abs(signal_strength) / latest_data['close'].iloc[-1] # LSTM Model lstm_features = scaled_features.reshape((scaled_features.shape[0], scaled_features.shape[1], 1)) lstm_prediction = self.models['lstm'].predict(lstm_features) signal_strength_lstm = lstm_prediction[-1][0] - latest_data['close'].iloc[-1] category_lstm = "strong buy" if signal_strength_lstm > 0.005 else "weak buy" if signal_strength_lstm > 0 else "strong sell" if signal_strength_lstm < -0.005 else "weak sell" signals['lstm'] = category_lstm confidences['lstm'] = abs(signal_strength_lstm) / latest_data['close'].iloc[-1] # Hybrid CNN-LSTM Model hybrid_prediction = self.models['hybrid_cnn_lstm'].predict(lstm_features) signal_strength_hybrid = hybrid_prediction[-1][0] - latest_data['close'].iloc[-1] category_hybrid = "strong buy" if signal_strength_hybrid > 0.005 else "weak buy" if signal_strength_hybrid > 0 else "strong sell" if signal_strength_hybrid < -0.005 else "weak sell" signals['hybrid_cnn_lstm'] = category_hybrid confidences['hybrid_cnn_lstm'] = abs(signal_strength_hybrid) / latest_data['close'].iloc[-1] # ARIMA Model arima_forecast = self.models['arima'].forecast(steps=1) signal_strength_arima = arima_forecast[0] - latest_data['close'].iloc[-1] category_arima = "strong buy" if signal_strength_arima > 0.005 else "weak buy" if signal_strength_arima > 0 else "strong sell" if signal_strength_arima < -0.005 else "weak sell" signals['arima'] = category_arima confidences['arima'] = abs(signal_strength_arima) / latest_data['close'].iloc[-1] # Prophet Model prophet_forecast = self.models['prophet'].predict(pd.DataFrame({'ds': [latest_data.index[-1] + pd.Timedelta(minutes=5)]})) signal_strength_prophet = prophet_forecast['yhat'].iloc[0] - latest_data['close'].iloc[-1] category_prophet = "strong buy" if signal_strength_prophet > 0.005 else "weak buy" if signal_strength_prophet > 0 else "strong sell" if signal_strength_prophet < -0.005 else "weak sell" signals['prophet'] = category_prophet confidences['prophet'] = abs(signal_strength_prophet) / latest_data['close'].iloc[-1] return signals, confidences async def concurrent_signal_generation(self, symbol: str): tasks = [] for timeframe in self.timeframes: data = await self.fetch_data(symbol, timeframe) preprocessed_data = self.preprocess_data(data) await self.train_models(preprocessed_data) task = self.generate_signals(preprocessed_data.iloc[-30:]) tasks.append(task) results = await asyncio.gather(*tasks) return results async def save_signals_to_db(self, symbol: str, timeframe: str, final_decision: str, signals: Dict[str, str], confidences: Dict[str, float]): Signal.objects.update_or_create( symbol=symbol, timeframe=timeframe, defaults={ 'decision': final_decision, 'signals': json.dumps(signals), 'confidences': json.dumps(confidences), } ) logger.info(f"Signals saved for {symbol} {timeframe}") async def run(self): await self.initialize() for symbol in self.symbols: results = await self.concurrent_signal_generation(symbol) for timeframe, (signals, confidences) in zip(self.timeframes, results): final_decision = "buy" if sum(1 for s in signals.values() if 'buy' in s) > len(signals) / 2 else "sell" await self.save_signals_to_db(symbol, timeframe, final_decision, signals, confidences) await self.session.close()if name == "main": symbols = ['BTC/USDT', 'ETH/USDT'] # Example list of symbols bot = MachineTradingBot(symbols) asyncio.run(bot.run())
Between the code above and the one below, wic one is likely to produce more accurate signals. What are the strengths and weaknesses of each and what would you recommend I use. Also, anything that I can do to improve it's accuracy further
import asyncio import numpy as np import pandas as pd from datetime import datetime, timedelta from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor from sklearn.preprocessing import StandardScaler, RobustScaler from sklearn.model_selection import TimeSeriesSplit from sklearn.metrics import mean_absolute_error, precision_score, recall_score, f1_score import lightgbm as lgb from xgboost import XGBRegressor import tensorflow as tf from tensorflow.keras.models import Sequential from tensorflow.keras.layers import LSTM, Dense, Dropout from tensorflow.keras.callbacks import EarlyStopping from tensorflow.keras.optimizers import Adam import logging from typing import Dict, List, Tuple, Optional from django.utils import timezone from django.db.models import Q from decimal import Decimal
Assume these are your Django models
from .models import PriceData, Signal, PerformanceMetrics
logging.basicConfig(level=logging.INFO) logger = logging.getLogger(name)
class EnhancedTradingBot: def init(self, symbols: List[str]): self.symbols = symbols self.timeframes = ['1m', '5m', '15m', '1h'] self.models = self._initialize_models() self.scalers = { 'price': RobustScaler(), 'volume': RobustScaler(), 'features': StandardScaler() } self.performance_metrics = {} self.volatility_windows = {'1m': 60, '5m': 24, '15m': 16, '1h': 24} self.min_success_rate = 65 # Minimum success rate to generate signals
def _initialize_models(self) -> Dict: return { 'random_forest': RandomForestRegressor( n_estimators=100, max_depth=10, min_samples_split=5, min_samples_leaf=2, n_jobs=-1, random_state=42 ), 'lightgbm': lgb.LGBMRegressor( n_estimators=100, learning_rate=0.05, max_depth=7, num_leaves=31, n_jobs=-1, random_state=42 ), 'xgboost': XGBRegressor( n_estimators=100, learning_rate=0.05, max_depth=7, n_jobs=-1, random_state=42 ), 'gradient_boosting': GradientBoostingRegressor( n_estimators=100, learning_rate=0.05, max_depth=5, random_state=42 ), 'lstm': None # Will be initialized during training } async def fetch_historical_data(self, symbol: str, timeframe: str) -> pd.DataFrame: """Fetch historical data from Django ORM""" queryset = PriceData.objects.filter( symbol=symbol, timeframe=timeframe ).order_by('timestamp') df = pd.DataFrame.from_records( queryset.values('timestamp', 'open', 'high', 'low', 'close', 'volume') ) df.set_index('timestamp', inplace=True) return df def calculate_dynamic_thresholds(self, data: pd.DataFrame, timeframe: str) -> Tuple[float, float]: """Calculate dynamic thresholds based on volatility""" returns = data['close'].pct_change() volatility = returns.rolling(window=self.volatility_windows[timeframe]).std() current_volatility = volatility.iloc[-1] # Adjust thresholds based on volatility base_threshold = 0.005 volatility_multiplier = current_volatility / returns.std() buy_threshold = base_threshold * volatility_multiplier sell_threshold = -base_threshold * volatility_multiplier return buy_threshold, sell_threshold def create_features(self, data: pd.DataFrame) -> pd.DataFrame: """Create advanced features for ML models""" df = data.copy() # Price-based features df['returns'] = df['close'].pct_change() df['log_returns'] = np.log1p(df['returns']) # Volatility features df['volatility'] = df['returns'].rolling(window=20).std() df['volatility_change'] = df['volatility'].pct_change() # Volume-based features df['volume_ma'] = df['volume'].rolling(window=20).mean() df['volume_std'] = df['volume'].rolling(window=20).std() df['volume_ratio'] = df['volume'] / df['volume_ma'] # Price momentum for window in [5, 10, 20, 30]: df[f'momentum_{window}'] = df['close'].pct_change(periods=window) df[f'volume_momentum_{window}'] = df['volume'].pct_change(periods=window) # Price differences df['price_range'] = df['high'] - df['low'] df['price_range_ratio'] = df['price_range'] / df['close'] # Target variable (future returns) df['target'] = df['close'].shift(-1) / df['close'] - 1 return df.dropna() def prepare_lstm_data(self, data: pd.DataFrame, lookback: int = 60) -> Tuple[np.ndarray, np.ndarray]: """Prepare data for LSTM model""" features = data.drop(['target'], axis=1).values target = data['target'].values X, y = [], [] for i in range(len(features) - lookback): X.append(features[i:(i + lookback)]) y.append(target[i + lookback]) return np.array(X), np.array(y) def create_lstm_model(self, input_shape: Tuple) -> Sequential: """Create LSTM model with optimized architecture""" model = Sequential([ LSTM(50, input_shape=input_shape, return_sequences=True), Dropout(0.2), LSTM(30, return_sequences=False), Dropout(0.2), Dense(1) ]) model.compile( optimizer=Adam(learning_rate=0.001), loss='mse', metrics=['mae'] ) return model async def train_models(self, data: pd.DataFrame, timeframe: str): """Train all models with the provided data""" features_df = self.create_features(data) # Split features and target X = features_df.drop(['target'], axis=1) y = features_df['target'] # Scale features X_scaled = self.scalers['features'].fit_transform(X) # Time series split tscv = TimeSeriesSplit(n_splits=5) # Train traditional ML models for name, model in self.models.items(): if name != 'lstm': logger.info(f"Training {name} for {timeframe}") model.fit(X_scaled, y) # Train LSTM X_lstm, y_lstm = self.prepare_lstm_data(features_df) self.models['lstm'] = self.create_lstm_model((X_lstm.shape[1], X_lstm.shape[2])) early_stopping = EarlyStopping( monitor='val_loss', patience=5, restore_best_weights=True ) self.models['lstm'].fit( X_lstm, y_lstm, epochs=50, batch_size=32, validation_split=0.2, callbacks=[early_stopping], verbose=0 ) def calculate_support_resistance(self, data: pd.DataFrame) -> Tuple[float, float]: """Calculate support and resistance levels""" price_range = data['high'].max() - data['low'].min() support = data['low'].iloc[-1] - (price_range * 0.1) resistance = data['high'].iloc[-1] + (price_range * 0.1) return support, resistance def calculate_success_rate(self, predictions: Dict[str, float], confidence_scores: Dict[str, float]) -> float: """Calculate success rate based on model predictions and confidence""" weighted_confidence = sum(pred * conf for pred, conf in zip(predictions.values(), confidence_scores.values())) return min(abs(weighted_confidence) * 100, 95) # Cap at 95% async def generate_signals(self, symbol: str, timeframe: str) -> Optional[Dict]: """Generate trading signals with comprehensive analysis""" try: # Fetch and prepare data data = await self.fetch_historical_data(symbol, timeframe) features_df = self.create_features(data) if len(features_df) < 100: # Minimum required data return None X = features_df.drop(['target'], axis=1).iloc[-1:] X_scaled = self.scalers['features'].transform(X) # Get predictions from all models predictions = {} confidence_scores = {} for name, model in self.models.items(): if name != 'lstm': pred = model.predict(X_scaled)[0] predictions[name] = pred confidence_scores[name] = abs(pred) else: X_lstm = self.prepare_lstm_data(features_df.iloc[-60:], lookback=60)[0] pred = model.predict(X_lstm)[-1][0] predictions[name] = pred confidence_scores[name] = abs(pred) # Calculate dynamic thresholds buy_threshold, sell_threshold = self.calculate_dynamic_thresholds(data, timeframe) # Calculate weighted prediction weights = { 'random_forest': 0.2, 'lightgbm': 0.25, 'xgboost': 0.25, 'gradient_boosting': 0.15, 'lstm': 0.15 } weighted_prediction = sum(pred * weights[name] for name, pred in predictions.items()) # Calculate success rate success_rate = self.calculate_success_rate(predictions, confidence_scores) if success_rate < self.min_success_rate: return None # Determine signal type signal_type = 'neutral' if weighted_prediction > buy_threshold: signal_type = 'strong_buy' if weighted_prediction > 2 * buy_threshold else 'buy' elif weighted_prediction < sell_threshold: signal_type = 'strong_sell' if weighted_prediction < 2 * sell_threshold else 'sell' # Calculate support and resistance support, resistance = self.calculate_support_resistance(data) # Prepare signal details current_price = float(data['close'].iloc[-1]) signal_details = { 'symbol': symbol, 'timeframe': timeframe, 'timestamp': timezone.now(), 'price_timestamp': data.index[-1], 'price': current_price, 'change_24h': float(data['close'].pct_change(periods=24).iloc[-1] * 100), 'signal': signal_type, 'stop_loss': support if signal_type in ['buy', 'strong_buy'] else resistance, 'take_profit': resistance if signal_type in ['buy', 'strong_buy'] else support, 'success_rate': success_rate, 'model_predictions': predictions, 'confidence_scores': confidence_scores, 'weighted_prediction': weighted_prediction, 'thresholds': { 'buy': buy_threshold, 'sell': sell_threshold } } return signal_details except Exception as e: logger.error(f"Error generating signal for {symbol} {timeframe}: {str(e)}") return None async def save_signal(self, signal_details: Dict): """Save signal to database""" try: Signal.objects.create( symbol=signal_details['symbol'], timeframe=signal_details['timeframe'], timestamp=signal_details['timestamp'], price_timestamp=signal_details['price_timestamp'], price=Decimal(str(signal_details['price'])), change_24h=Decimal(str(signal_details['change_24h'])), signal_type=signal_details['signal'], stop_loss=Decimal(str(signal_details['stop_loss'])), take_profit=Decimal(str(signal_details['take_profit'])), success_rate=Decimal(str(signal_details['success_rate'])), details=signal_details ) except Exception as e: logger.error(f"Error saving signal: {str(e)}") async def run(self): """Main execution loop""" try: tasks = [] for symbol in self.symbols: for timeframe in self.timeframes: # Train models data = await self.fetch_historical_data(symbol, timeframe) await self.train_models(data, timeframe) # Generate and save signals tasks.append(self.generate_signals(symbol, timeframe)) signals = await asyncio.gather(*tasks) # Save valid signals save_tasks = [] for signal in signals: if signal: save_tasks.append(self.save_signal(signal)) if save_tasks: await asyncio.gather(*tasks) except Exception as e: logger.error(f"Error in main execution loop: {str(e)}")if name == "main": symbols = ['BTC/USDT', 'ETH/USDT'] # Example symbols bot = EnhancedTradingBot(symbols) asyncio.run(bot.run())
- Assistant
Comparing the two trading bots, the "MachineTradingBot" and the "EnhancedTradingBot," here are some strengths, weaknesses, and potential recommendations:
Strengths and Weaknesses
MachineTradingBot:
-
Strengths:
- Diverse Model Collection: Utilizes a wide variety of models, including Prophet, ARIMA, and hybrid models like CNN-LSTM.
- Technical Indicators and Fourier Features: Incorporates technical indicators and Fourier features to offer comprehensive feature representation.
- Concurrency: Utilizes asyncio for concurrent operations, enhancing performance for real-time data processing.
-
Weaknesses:
- Complexity: High complexity due to many different models and features, which could lead to overfitting.
- Signal Categorization: Uses arbitrary thresholds that may not adapt well to changing market conditions.
- Scalability Issues: Due to the number of models, scalability can become an issue as more symbols or timeframes are added.
EnhancedTradingBot:
-
Strengths:
- Dynamic Thresholds: Implements dynamic thresholds based on volatility, making it adaptive to different market conditions.
- Advanced Feature Generation: Focuses on sophisticated feature engineering including momentum and volatility features.
- Success Rate Filtering: Has a success rate threshold to filter out low-confidence predictions, potentially improving signal quality.
-
Weaknesses:
- Heavy Dependency on Static Weights: Uses fixed weights for model prediction aggregation which might not always capture the optimal combination.
- Limited Model Range: Fewer model types, potentially missing out on patterns that more diverse models could capture.
- Single Approach on Model Training: Relies on traditional time series split without consideration for more modern cross-validation techniques.
Recommendations
-
Model Diversification: Consider combining elements of both approaches for a hybrid solution, utilizing model diversity from the MachineTradingBot with the adaptive mechanisms from the EnhancedTradingBot.
-
Adaptive Weighting: Implement an adaptive weighting scheme for models based on past performance to dynamically allocate weights.
-
Hyperparameter Optimization: Both models might benefit from a more systematic hyperparameter optimization using Bayesian optimization techniques rather than static or randomized search.
-
Real-Time Adaptation: Enrich real-time adaptability by integrating reinforcement learning strategies to continuously learn and adapt to the trading environment.
-
Backtesting and Validation: Rigorously backtest both models using historical data, evaluating their accuracy and robustness under different market conditions to choose the more suitable one.
In conclusion, both models have their unique advantages and could be improved by integrating their best parts. If adapting just one, the EnhancedTradingBot could be more advantageous due to its dynamic features and adaptability, with further enhancements on model diversity and weighting schemes.
-
- User
Between EnhancedTradingBot above and the one below. Wic is better import asyncio import numpy as np import pandas as pd from datetime import datetime, timedelta from sklearn.ensemble import RandomForestRegressor from sklearn.preprocessing import RobustScaler from sklearn.model_selection import TimeSeriesSplit from sklearn.metrics import mean_absolute_error import lightgbm as lgb from tensorflow.keras.models import Sequential from tensorflow.keras.layers import LSTM, Dense, Dropout from tensorflow.keras.callbacks import EarlyStopping from tensorflow.keras.optimizers import Adam import optuna import logging from typing import Dict, List, Optional from django.utils import timezone from django.db.models import Q from decimal import Decimal import psutil import gc
Assume these are your Django models
from .models import PriceData, Signal, PerformanceMetrics
logging.basicConfig(level=logging.INFO) logger = logging.getLogger(name)
class OptimizedTradingBot: def init(self, batch_size: int = 10): self.batch_size = batch_size # Process symbols in batches self.timeframes = ['15m', '1h'] # Reduced timeframes for better resource management self.models = {} self.scaler = RobustScaler() # Single scaler for simplicity self.performance_metrics = {} self.volatility_windows = {'15m': 16, '1h': 24} self.min_success_rate = 70 # Increased for better signal quality self.study_storage = {} self.best_params = {}
# Resource management total_ram = psutil.virtual_memory().total self.memory_limit = total_ram * 0.6 # 60% of total RAM self.cpu_cores = min(psutil.cpu_count(logical=False), 4) # Limited to 4 cores # Model selection (reduced to 3 most efficient models) self.model_types = ['random_forest', 'lightgbm', 'lstm'] async def fetch_symbols(self) -> List[str]: """Fetch available trading pairs from database""" try: # Assuming PriceData model has a symbols field or method symbols = await PriceData.objects.values_list('symbol', flat=True).distinct() return list(symbols) except Exception as e: logger.error(f"Error fetching symbols: {str(e)}") return ['BTC/USDT', 'ETH/USDT'] # Fallback symbols async def fetch_historical_data(self, symbol: str, timeframe: str, lookback_periods: int = 1000) -> pd.DataFrame: """Fetch historical price data with error handling and validation""" try: end_time = timezone.now() start_time = end_time - timedelta( hours=lookback_periods if timeframe == '1h' else lookback_periods//4 ) query = Q(symbol=symbol) & Q(timeframe=timeframe) & \ Q(timestamp__range=(start_time, end_time)) data = await PriceData.objects.filter(query).order_by('timestamp').values() df = pd.DataFrame(data) if len(df) < lookback_periods * 0.9: # Ensure sufficient data logger.warning(f"Insufficient data for {symbol} {timeframe}") return None return df except Exception as e: logger.error(f"Error fetching historical data: {str(e)}") return None def create_features(self, df: pd.DataFrame) -> pd.DataFrame: """Create enhanced technical features""" df = df.copy() # Price-based features df['returns'] = df['close'].pct_change() df['log_returns'] = np.log1p(df['returns']) # Volatility features df['volatility'] = df['returns'].rolling(window=20).std() df['volatility_ratio'] = df['volatility'] / df['volatility'].rolling(window=100).mean() # Volume features df['volume_ma'] = df['volume'].rolling(window=20).mean() df['volume_ratio'] = df['volume'] / df['volume_ma'] # Price momentum df['momentum'] = df['close'].pct_change(periods=12) df['acceleration'] = df['momentum'].diff() # Trend features df['ma_20'] = df['close'].rolling(window=20).mean() df['ma_50'] = df['close'].rolling(window=50).mean() df['trend'] = (df['ma_20'] / df['ma_50']) - 1 # Target variable (future returns) df['target'] = df['returns'].shift(-1) # Clean up df = df.dropna() return df def prepare_lstm_data(self, df: pd.DataFrame, sequence_length: int = 20) -> tuple: """Prepare data for LSTM model with memory efficiency""" features = df.drop(['target'], axis=1).values target = df['target'].values X, y = [], [] for i in range(len(features) - sequence_length): X.append(features[i:i + sequence_length]) y.append(target[i + sequence_length]) return np.array(X), np.array(y) async def optimize_random_forest(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize RandomForest with reduced parameter space""" params = { 'n_estimators': trial.suggest_int('rf_n_estimators', 50, 100), 'max_depth': trial.suggest_int('rf_max_depth', 3, 10), 'min_samples_split': trial.suggest_int('rf_min_samples_split', 2, 8) } model = RandomForestRegressor(**params, n_jobs=self.cpu_cores, random_state=42) return await self._evaluate_model(model, X, y) async def optimize_lightgbm(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize LightGBM with memory-efficient parameters""" params = { 'n_estimators': trial.suggest_int('lgb_n_estimators', 50, 100), 'max_depth': trial.suggest_int('lgb_max_depth', 3, 8), 'learning_rate': trial.suggest_float('lgb_learning_rate', 1e-3, 0.1, log=True), 'num_leaves': trial.suggest_int('lgb_num_leaves', 15, 31) } model = lgb.LGBMRegressor(**params, n_jobs=self.cpu_cores, random_state=42) return await self._evaluate_model(model, X, y) async def optimize_lstm(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize LSTM with reduced complexity""" params = { 'units': trial.suggest_int('lstm_units', 32, 64), 'dropout': trial.suggest_float('lstm_dropout', 0.1, 0.3), 'learning_rate': trial.suggest_float('lstm_learning_rate', 1e-3, 1e-2, log=True), 'batch_size': 32 # Fixed batch size for stability } model = Sequential([ LSTM(params['units'], input_shape=(X.shape[1], X.shape[2])), Dropout(params['dropout']), Dense(1) ]) model.compile( optimizer=Adam(learning_rate=params['learning_rate']), loss='mse' ) early_stopping = EarlyStopping(monitor='val_loss', patience=3) tscv = TimeSeriesSplit(n_splits=3) scores = [] for train_idx, val_idx in tscv.split(X): history = model.fit( X[train_idx], y[train_idx], epochs=20, # Reduced epochs batch_size=params['batch_size'], validation_data=(X[val_idx], y[val_idx]), callbacks=[early_stopping], verbose=0 ) scores.append(min(history.history['val_loss'])) return np.mean(scores) async def _evaluate_model(self, model, X: np.ndarray, y: np.ndarray) -> float: """Evaluate model performance with cross-validation""" tscv = TimeSeriesSplit(n_splits=3) scores = [] for train_idx, val_idx in tscv.split(X): X_train, X_val = X[train_idx], X[val_idx] y_train, y_val = y[train_idx], y[val_idx] model.fit(X_train, y_train) pred = model.predict(X_val) score = mean_absolute_error(y_val, pred) scores.append(score) return np.mean(scores) async def optimize_models(self, data: pd.DataFrame, timeframe: str): """Run optimization with resource constraints""" features_df = self.create_features(data) X = features_df.drop(['target'], axis=1) y = features_df['target'] X_scaled = self.scaler.fit_transform(X) X_lstm, y_lstm = self.prepare_lstm_data(features_df) n_trials = 30 # Reduced number of trials for model_type in self.model_types: if psutil.virtual_memory().available < self.memory_limit: logger.warning("Low memory. Cleaning up...") gc.collect() study = optuna.create_study(direction='minimize') if model_type == 'lstm': await self._run_optimization(study, model_type, X_lstm, y_lstm, n_trials) else: await self._run_optimization(study, model_type, X_scaled, y, n_trials) self.best_params[f"{timeframe}_{model_type}"] = study.best_params self.study_storage[f"{timeframe}_{model_type}"] = study async def _run_optimization(self, study, model_type: str, X: np.ndarray, y: np.ndarray, n_trials: int): """Run optimization for a single model""" optimize_func = getattr(self, f'optimize_{model_type}') study.optimize( lambda trial: asyncio.run(optimize_func(trial, X, y)), n_trials=n_trials ) async def generate_signals(self, symbol: str, timeframe: str) -> Optional[Dict]: """Generate trading signals with enhanced validation""" try: data = await self.fetch_historical_data(symbol, timeframe) if data is None: return None features_df = self.create_features(data) X = features_df.drop(['target'], axis=1) X_scaled = self.scaler.transform(X) predictions = [] weights = {'random_forest': 0.4, 'lightgbm': 0.4, 'lstm': 0.2} for model_type in self.model_types: if model_type == 'lstm': X_lstm, _ = self.prepare_lstm_data(features_df) pred = self.models[model_type].predict(X_lstm[-1:]) else: pred = self.models[model_type].predict(X_scaled[-1:]) predictions.append(pred[0] * weights[model_type]) ensemble_prediction = sum(predictions) # Generate signal with enhanced validation signal = self._validate_and_create_signal( symbol, timeframe, ensemble_prediction, data ) return signal except Exception as e: logger.error(f"Error generating signals: {str(e)}") return None def _validate_and_create_signal(self, symbol: str, timeframe: str, prediction: float, data: pd.DataFrame) -> Optional[Dict]: """Validate and create trading signal with risk management""" try: current_price = data['close'].iloc[-1] volatility = data['close'].pct_change().std() * np.sqrt(252) # Dynamic threshold based on volatility threshold = max(0.001, volatility * 0.1) if abs(prediction) < threshold: return None signal_type = 'buy' if prediction > 0 else 'sell' # Dynamic stop loss and take profit based on volatility stop_loss = current_price * (1 - volatility) if signal_type == 'buy' else current_price * (1 + volatility) take_profit = current_price * (1 + volatility * 1.5) if signal_type == 'buy' else current_price * (1 - volatility * 1.5) # Calculate success rate based on historical predictions success_rate = self._calculate_success_rate(data, prediction) if success_rate < self.min_success_rate: return None return { 'symbol': symbol, 'timeframe': timeframe, 'timestamp': timezone.now(), 'price_timestamp': data.index[-1], 'price': current_price, 'change_24h': data['close'].pct_change(24).iloc[-1], 'signal': signal_type, 'stop_loss': stop_loss, 'take_profit': take_profit, 'success_rate': success_rate, 'volatility': volatility } except Exception as e: logger.error(f"Error validating signal: {str(e)}") return Nonedef _calculate_success_rate(self, data: pd.DataFrame, current_prediction: float) -> float: """Calculate success rate of previous predictions""" try: predictions = [] actuals = []
for i in range(len(data) - 20, len(data)): features = self.create_features(data.iloc[:i]) X = features.drop(['target'], axis=1) X_scaled = self.scaler.transform(X) model_predictions = [] for model_type in self.model_types: if model_type == 'lstm': X_lstm, _ = self.prepare_lstm_data(features) pred = self.models[model_type].predict(X_lstm[-1:]) else: pred = self.models[model_type].predict(X_scaled[-1:]) model_predictions.append(pred[0]) avg_pred = np.mean(model_predictions) predictions.append(avg_pred > 0) actuals.append(data['returns'].iloc[i] > 0) return np.mean(np.array(predictions) == np.array(actuals)) * 100 except Exception as e: logger.error(f"Error calculating success rate: {str(e)}") return 0.0 async def save_signal(self, signal_details: Dict): """Save signal to database with enhanced error handling""" try: if not signal_details: return signal = Signal.objects.create( symbol=signal_details['symbol'], timeframe=signal_details['timeframe'], timestamp=signal_details['timestamp'], price_timestamp=signal_details['price_timestamp'], price=Decimal(str(signal_details['price'])), change_24h=Decimal(str(signal_details['change_24h'])), signal_type=signal_details['signal'], stop_loss=Decimal(str(signal_details['stop_loss'])), take_profit=Decimal(str(signal_details['take_profit'])), success_rate=Decimal(str(signal_details['success_rate'])), analysis_type='ml_ensemble', details={ 'models_used': self.model_types, 'volatility': float(signal_details['volatility']), 'optimization_metrics': { model: str(study.best_value) for model, study in self.study_storage.items() if model.startswith(signal_details['timeframe']) } } ) # Update performance metrics await self._update_performance_metrics(signal_details['symbol'], signal_details['timeframe']) return signal except Exception as e: logger.error(f"Error saving signal: {str(e)}") return None async def _update_performance_metrics(self, symbol: str, timeframe: str): """Update performance metrics for monitoring""" try: # Get recent signals recent_signals = Signal.objects.filter( symbol=symbol, timeframe=timeframe, timestamp__gte=timezone.now() - timedelta(days=7) ) # Calculate metrics total_signals = len(recent_signals) successful_signals = len([s for s in recent_signals if s.is_successful()]) if total_signals > 0: success_rate = (successful_signals / total_signals) * 100 else: success_rate = 0 # Update or create metrics PerformanceMetrics.objects.update_or_create( symbol=symbol, timeframe=timeframe, defaults={ 'success_rate': Decimal(str(success_rate)), 'total_signals': total_signals, 'last_updated': timezone.now() } ) except Exception as e: logger.error(f"Error updating performance metrics: {str(e)}") async def process_symbol_batch(self, symbols: List[str]): """Process a batch of symbols""" try: for timeframe in self.timeframes: for symbol in symbols: if psutil.virtual_memory().available < self.memory_limit: logger.warning("Low memory. Cleaning up...") gc.collect() await asyncio.sleep(1) # Give system time to recover # Fetch and prepare data data = await self.fetch_historical_data(symbol, timeframe) if data is not None: # Train models if needed await self.optimize_models(data, timeframe) # Generate and save signals signal = await self.generate_signals(symbol, timeframe) if signal: await self.save_signal(signal) await asyncio.sleep(0.1) # Prevent overwhelming the system except Exception as e: logger.error(f"Error processing symbol batch: {str(e)}") async def run(self): """Main execution loop with enhanced error handling and resource management""" try: # Fetch available symbols symbols = await self.fetch_symbols() if not symbols: logger.error("No symbols available for trading") return # Process symbols in batches symbol_batches = [symbols[i:i + self.batch_size] for i in range(0, len(symbols), self.batch_size)] for batch in symbol_batches: await self.process_symbol_batch(batch) # Memory management after each batch gc.collect() await asyncio.sleep(1) logger.info("Trading bot execution completed successfully") except Exception as e: logger.error(f"Error in main execution loop: {str(e)}") finally: # Cleanup gc.collect() async def start(self): """Start the trading bot with error handling""" try: logger.info("Starting trading bot...") while True: await self.run() # Wait before next iteration await asyncio.sleep(300) # 5 minutes between runs except KeyboardInterrupt: logger.info("Trading bot stopped by user") except Exception as e: logger.error(f"Critical error in trading bot: {str(e)}") finally: logger.info("Trading bot shutdown complete")if name == "main": # Initialize and run the bot bot = OptimizedTradingBot(batch_size=5) # Process 5 symbols at a time asyncio.run(bot.start())
- Assistant
When comparing the
EnhancedTradingBotand theOptimizedTradingBot, it's essential to evaluate their strengths and weaknesses to determine which might be more suitable for producing accurate trading signals. Here are some considerations:Similarities
- Data Handling: Both bots handle historical data fetching and feature engineering similarly.
- Model Usage: Both utilize models like
RandomForestRegressor,LightGBM, and anLSTM. - Signal Generation: Both generate signals by aggregating predictions from multiple models.
Differences
EnhancedTradingBot:
-
Strengths:
- Dynamic Thresholds: Implements dynamic thresholds based on volatility for signal determination, allowing the bot to adapt to market conditions.
- Rich Feature Set: Extensive use of sophisticated features such as momentum, volatility, and trends.
- Success Rate Filtering: Uses a success rate threshold for filtering signals, which can improve the quality of signals.
-
Weaknesses:
- Resource Utilization: May not be optimized for resource use and scalability.
- Static Model Weights: Uses fixed weights in model predictions aggregation, which could reduce adaptability.
OptimizedTradingBot:
-
Strengths:
- Resource Management: Limits the use of system resources (CPU and RAM) and includes garbage collection to enhance performance efficiency.
- Batch Processing: Processes symbols in batches to optimize resource usage and system performance.
- Model Optimization: Employs
optunafor hyperparameter tuning, potentially leading to more accurate models. - Fewer Timeframes: Maintains focus on fewer timeframes to ensure better allocation of computational resources.
-
Weaknesses:
- Reduced Model Complexity: Limits feature complexity and the number of models, which might reduce identification of complex patterns.
- Simplified Training: Reduction in trials and epochs for efficiency might lead to underfitting.
Recommendations
Based on the features and design choices:
- For Resource-Constrained Environments: The
OptimizedTradingBotis likely a better choice due to its built-in memory and CPU management features. The use ofoptunafor efficient hyperparameter tuning can potentially improve model performance despite fewer trials and simpler architecture. - For Rich Feature Analysis and Market Adaptation: The
EnhancedTradingBotwould be more appropriate, especially with its dynamic thresholds and extensive feature engineering for adapting to market conditions.
Improvement Suggestions
- Hybrid Approach: Consider incorporating the enhanced dynamic threshold and feature engineering techniques from
EnhancedTradingBotinto theOptimizedTradingBotfor better market adaptability. - Adaptive Weighting: Implement a more dynamic approach to model weighting based on real-time performance metrics or context-aware decisions.
- Comprehensive Backtesting: Conduct thorough backtesting under various market conditions using historical data to validate performance and improve predictive accuracy.
Ultimately, if system resource constraints are a significant concern, then focusing on the
OptimizedTradingBotwith improvements could offer a balance between precision and performance efficiency. - User
I made other changes, how is it now? import asyncio import numpy as np import pandas as pd from datetime import datetime, timedelta from sklearn.ensemble import RandomForestRegressor from sklearn.preprocessing import RobustScaler from sklearn.model_selection import TimeSeriesSplit from sklearn.metrics import mean_absolute_error import lightgbm as lgb from keras import Sequential from tensorflow.python.keras.layers.recurrent import LSTM from tensorflow.python.keras.layers import Dense, Dropout, Attention, Conv1D from keras.src.callbacks import EarlyStopping from tensorflow.python.keras.optimizer_v1 import Adam from statsmodels.tsa.arima.model import ARIMA from xgboost import XGBRegressor import optuna import logging from typing import Dict, List, Optional from django.utils import timezone from django.db.models import Q from decimal import Decimal import psutil import gc
Assume these are your Django models
from .models import PriceData, Signal, PerformanceMetrics
logging.basicConfig(level=logging.INFO) logger = logging.getLogger(name)
class OptimizedTradingBot: def init(self, batch_size: int = 8): self.batch_size = batch_size self.timeframes = ['15m', '1h'] self.models = {} self.scaler = RobustScaler() self.performance_metrics = {} self.volatility_windows = {'15m': 16, '1h': 24} self.min_success_rate = 65 # Lowered for ensemble diversity self.study_storage = {} self.best_params = {}
# Resource management for 8GB RAM total_ram = psutil.virtual_memory().total self.memory_limit = total_ram * 0.5 # 50% of total RAM self.cpu_cores = min(psutil.cpu_count(logical=False), 2) # Limited to 2 cores for stability # Enhanced model selection (5 diverse models) self.model_types = [ 'random_forest', # Traditional ML 'xgboost', # Traditional ML 'lightgbm', # Traditional ML 'lstm', # Deep Learning 'arima' # Time Series ] # Model weights for ensemble self.model_weights = { 'random_forest': 0.25, 'xgboost': 0.20, 'lightgbm': 0.25, 'lstm': 0.15, 'arima': 0.15 } async def fetch_symbols(self) -> List[str]: """Fetch available trading pairs from database""" try: # Assuming PriceData model has a symbols field or method symbols = await PriceData.objects.values_list('symbol', flat=True).distinct() return list(symbols) except Exception as e: logger.error(f"Error fetching symbols: {str(e)}") return ['BTC/USDT', 'ETH/USDT'] # Fallback symbols async def fetch_historical_data(self, symbol: str, timeframe: str, lookback_periods: int = 1000) -> pd.DataFrame: """Fetch historical price data with error handling and validation""" try: end_time = timezone.now() start_time = end_time - timedelta( hours=lookback_periods if timeframe == '1h' else lookback_periods//4 ) query = Q(symbol=symbol) & Q(timeframe=timeframe) & \ Q(timestamp__range=(start_time, end_time)) data = await PriceData.objects.filter(query).order_by('timestamp').values() df = pd.DataFrame(data) if len(df) < lookback_periods * 0.9: # Ensure sufficient data logger.warning(f"Insufficient data for {symbol} {timeframe}") return None return df except Exception as e: logger.error(f"Error fetching historical data: {str(e)}") return None def create_features(self, df: pd.DataFrame) -> pd.DataFrame: """Create enhanced technical features""" df = df.copy() # Price-based features df['returns'] = df['close'].pct_change() df['log_returns'] = np.log1p(df['returns']) # Volatility features df['volatility'] = df['returns'].rolling(window=20).std() df['volatility_ratio'] = df['volatility'] / df['volatility'].rolling(window=100).mean() # Volume features df['volume_ma'] = df['volume'].rolling(window=20).mean() df['volume_ratio'] = df['volume'] / df['volume_ma'] # Price momentum df['momentum'] = df['close'].pct_change(periods=12) df['acceleration'] = df['momentum'].diff() # Trend features df['ma_20'] = df['close'].rolling(window=20).mean() df['ma_50'] = df['close'].rolling(window=50).mean() df['trend'] = (df['ma_20'] / df['ma_50']) - 1 # Target variable (future returns) df['target'] = df['returns'].shift(-1) # Clean up df = df.dropna() return df def prepare_lstm_data(self, df: pd.DataFrame, sequence_length: int = 20) -> tuple: """Prepare data for LSTM model with memory efficiency""" features = df.drop(['target'], axis=1).values target = df['target'].values X, y = [], [] for i in range(len(features) - sequence_length): X.append(features[i:i + sequence_length]) y.append(target[i + sequence_length]) return np.array(X), np.array(y) async def optimize_random_forest(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize RandomForest with reduced parameter space""" params = { 'n_estimators': trial.suggest_int('rf_n_estimators', 50, 100), 'max_depth': trial.suggest_int('rf_max_depth', 3, 10), 'min_samples_split': trial.suggest_int('rf_min_samples_split', 2, 8) } model = RandomForestRegressor(**params, n_jobs=self.cpu_cores, random_state=42) return await self._evaluate_model(model, X, y) async def optimize_lightgbm(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize LightGBM with memory-efficient parameters""" params = { 'n_estimators': trial.suggest_int('lgb_n_estimators', 50, 100), 'max_depth': trial.suggest_int('lgb_max_depth', 3, 8), 'learning_rate': trial.suggest_float('lgb_learning_rate', 1e-3, 0.1, log=True), 'num_leaves': trial.suggest_int('lgb_num_leaves', 15, 31) } model = lgb.LGBMRegressor(**params, n_jobs=self.cpu_cores, random_state=42) return await self._evaluate_model(model, X, y) async def optimize_lstm(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize LSTM with reduced complexity""" params = { 'units': trial.suggest_int('lstm_units', 32, 64), 'dropout': trial.suggest_float('lstm_dropout', 0.1, 0.3), 'learning_rate': trial.suggest_float('lstm_learning_rate', 1e-3, 1e-2, log=True), 'batch_size': 32 # Fixed batch size for stability } model = Sequential([ LSTM(params['units'], input_shape=(X.shape[1], X.shape[2])), Dropout(params['dropout']), Dense(1) ]) model.compile( optimizer=Adam(learning_rate=params['learning_rate']), loss='mse' ) early_stopping = EarlyStopping(monitor='val_loss', patience=3) tscv = TimeSeriesSplit(n_splits=3) scores = [] for train_idx, val_idx in tscv.split(X): history = model.fit( X[train_idx], y[train_idx], epochs=20, # Reduced epochs batch_size=params['batch_size'], validation_data=(X[val_idx], y[val_idx]), callbacks=[early_stopping], verbose=0 ) scores.append(min(history.history['val_loss'])) return np.mean(scores) async def optimize_xgboost(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize XGBoost with memory-efficient parameters""" params = { 'n_estimators': trial.suggest_int('xgb_n_estimators', 50, 100), 'max_depth': trial.suggest_int('xgb_max_depth', 3, 8), 'learning_rate': trial.suggest_float('xgb_learning_rate', 1e-3, 0.1, log=True), 'subsample': trial.suggest_float('xgb_subsample', 0.6, 0.9), 'colsample_bytree': trial.suggest_float('xgb_colsample_bytree', 0.6, 0.9) } model = XGBRegressor( **params, n_jobs=self.cpu_cores, random_state=42, tree_method='hist', # Memory efficient histogram-based algorithm gpu_id=None # Disable GPU ) return await self._evaluate_model(model, X, y) async def optimize_arima(self, trial: optuna.Trial, X: np.ndarray, y: np.ndarray) -> float: """Optimize ARIMA parameters""" params = { 'p': trial.suggest_int('arima_p', 1, 3), 'd': trial.suggest_int('arima_d', 0, 1), 'q': trial.suggest_int('arima_q', 1, 3) } # Use only the last 500 observations for ARIMA to save memory y_subset = y[-500:] try: model = ARIMA(y_subset, order=(params['p'], params['d'], params['q'])) fitted = model.fit() pred = fitted.forecast(steps=1) score = mean_absolute_error([y_subset[-1]], [pred[0]]) return score except: return float('inf') async def optimize_models(self, data: pd.DataFrame, timeframe: str): """Run optimization with enhanced resource management""" features_df = self.create_features(data) X = features_df.drop(['target'], axis=1) y = features_df['target'] X_scaled = self.scaler.fit_transform(X) X_lstm, y_lstm = self.prepare_lstm_data(features_df) n_trials = 20 # Further reduced trials for memory efficiency for model_type in self.model_types: # Memory cleanup before each model optimization if psutil.virtual_memory().available < self.memory_limit: logger.warning("Low memory. Cleaning up...") gc.collect() await asyncio.sleep(2) # Give more time for memory cleanup study = optuna.create_study(direction='minimize') if model_type == 'lstm': await self._run_optimization(study, model_type, X_lstm, y_lstm, n_trials) elif model_type == 'arima': await self._run_optimization(study, model_type, None, y, n_trials) else: await self._run_optimization(study, model_type, X_scaled, y, n_trials) self.best_params[f"{timeframe}_{model_type}"] = study.best_params self.study_storage[f"{timeframe}_{model_type}"] = study async def _evaluate_model(self, model, X: np.ndarray, y: np.ndarray) -> float: """Evaluate model performance with cross-validation""" tscv = TimeSeriesSplit(n_splits=5) scores = [] for train_idx, val_idx in tscv.split(X): X_train, X_val = X[train_idx], X[val_idx] y_train, y_val = y[train_idx], y[val_idx] model.fit(X_train, y_train) pred = model.predict(X_val) score = mean_absolute_error(y_val, pred) scores.append(score) return np.mean(scores) async def _run_optimization(self, study, model_type: str, X: np.ndarray, y: np.ndarray, n_trials: int): """Run optimization for a single model""" optimize_func = getattr(self, f'optimize_{model_type}') study.optimize( lambda trial: asyncio.run(optimize_func(trial, X, y)), n_trials=n_trials ) async def generate_signals(self, symbol: str, timeframe: str) -> Optional[Dict]: """Generate trading signals using all 5 models in the ensemble""" try: data = await self.fetch_historical_data(symbol, timeframe) if data is None: return None features_df = self.create_features(data) X = features_df.drop(['target'], axis=1) X_scaled = self.scaler.transform(X) predictions = [] weights = [] # Process each model type for model_type in self.model_types: try: if psutil.virtual_memory().available < self.memory_limit: logger.warning("Low memory during prediction. Cleaning up...") gc.collect() await asyncio.sleep(0.5) if model_type == 'lstm': # LSTM requires sequence data X_lstm, _ = self.prepare_lstm_data(features_df) pred = self.models[model_type].predict(X_lstm[-1:]) pred_value = pred[0] elif model_type == 'arima': # ARIMA works directly with the close prices # Use only recent data for memory efficiency recent_close = data['close'].iloc[-500:] # Last 500 points model = ARIMA( recent_close, order=( self.best_params[f"{timeframe}_arima"]['arima_p'], self.best_params[f"{timeframe}_arima"]['arima_d'], self.best_params[f"{timeframe}_arima"]['arima_q'] ) ) fitted = model.fit() pred = fitted.forecast(steps=1) # Convert ARIMA prediction to return format current_price = recent_close.iloc[-1] pred_value = (pred[0] - current_price) / current_price elif model_type == 'random_forest': # Random Forest prediction pred = self.models[model_type].predict(X_scaled[-1:]) pred_value = pred[0] elif model_type == 'xgboost': # XGBoost prediction pred = self.models[model_type].predict(X_scaled[-1:]) pred_value = pred[0] elif model_type == 'lightgbm': # LightGBM prediction pred = self.models[model_type].predict(X_scaled[-1:]) pred_value = pred[0] # Add prediction and weight if valid if not np.isnan(pred_value) and not np.isinf(pred_value): predictions.append(pred_value * self.model_weights[model_type]) weights.append(self.model_weights[model_type]) except Exception as e: logger.error(f"Error in model {model_type}: {str(e)}") continue # Check if we have enough valid predictions if len(predictions) < 3: # Require at least 3 valid models logger.warning("Insufficient valid predictions") return None # Normalize weights for available predictions total_weight = sum(weights) if total_weight > 0: predictions = [p * (1/total_weight) for p in predictions] # Calculate ensemble prediction ensemble_prediction = sum(predictions) # Enhanced signal validation signal = await self._validate_and_create_signal( symbol, timeframe, ensemble_prediction, data ) # Add model contribution details to signal if signal: signal['model_contributions'] = { model_type: weight for model_type, weight in zip(self.model_types, weights) if weight > 0 } return signal except Exception as e: logger.error(f"Error generating signals: {str(e)}") return None async def save_signal(self, signal_details: Dict): """Save signal to database with enhanced model contribution tracking""" try: if not signal_details: return # Extract model contributions before removing from signal details model_contributions = signal_details.pop('model_contributions', {}) signal = Signal.objects.create( symbol=signal_details['symbol'], timeframe=signal_details['timeframe'], timestamp=signal_details['timestamp'], price_timestamp=signal_details['price_timestamp'], price=Decimal(str(signal_details['price'])), change_24h=Decimal(str(signal_details['change_24h'])), signal_type=signal_details['signal'], stop_loss=Decimal(str(signal_details['stop_loss'])), take_profit=Decimal(str(signal_details['take_profit'])), success_rate=Decimal(str(signal_details['success_rate'])), analysis_type='ml_ensemble', details={ 'models_used': self.model_types, 'model_contributions': model_contributions, 'volatility': float(signal_details['volatility']), 'volume_score': float(signal_details['volume_score']), 'trend_score': float(signal_details['trend_score']), 'atr': float(signal_details['atr']), 'optimization_metrics': { model: str(study.best_value) for model, study in self.study_storage.items() if model.startswith(signal_details['timeframe']) } } ) # Update performance metrics await self._update_performance_metrics(signal_details['symbol'], signal_details['timeframe']) return signal except Exception as e: logger.error(f"Error saving signal: {str(e)}") return None async def _validate_and_create_signal(self, symbol: str, timeframe: str, prediction: float, data: pd.DataFrame) -> Optional[Dict]: """Enhanced signal validation with additional risk metrics""" try: current_price = data['close'].iloc[-1] # Enhanced volatility calculation using different timeframes short_vol = data['close'].pct_change().rolling(window=12).std() * np.sqrt(252) long_vol = data['close'].pct_change().rolling(window=26).std() * np.sqrt(252) vol_ratio = short_vol.iloc[-1] / long_vol.iloc[-1] # Dynamic threshold based on volatility ratio threshold = max(0.001, short_vol.iloc[-1] * 0.15 * vol_ratio) if abs(prediction) < threshold: return None signal_type = 'buy' if prediction > 0 else 'sell' # Enhanced stop loss and take profit calculations atr = self._calculate_atr(data) stop_loss = current_price * (1 - atr * 2) if signal_type == 'buy' else current_price * (1 + atr * 2) take_profit = current_price * (1 + atr * 3) if signal_type == 'buy' else current_price * (1 - atr * 3) # Calculate confidence score success_rate = self._calculate_success_rate(data, prediction) volume_score = self._calculate_volume_score(data) trend_score = self._calculate_trend_score(data) confidence_score = (success_rate * 0.4 + volume_score * 0.3 + trend_score * 0.3) if confidence_score < self.min_success_rate: return None return { 'symbol': symbol, 'timeframe': timeframe, 'timestamp': timezone.now(), 'price_timestamp': data.index[-1], 'price': current_price, 'change_24h': data['close'].pct_change(24).iloc[-1], 'signal': signal_type, 'stop_loss': stop_loss, 'take_profit': take_profit, 'success_rate': confidence_score, 'volatility': short_vol.iloc[-1], 'volume_score': volume_score, 'trend_score': trend_score, 'atr': atr } except Exception as e: logger.error(f"Error validating signal: {str(e)}") return None def _calculate_atr(self, data: pd.DataFrame, period: int = 14) -> float: """Calculate Average True Range""" try: high = data['high'] low = data['low'] close = data['close'] tr1 = high - low tr2 = abs(high - close.shift()) tr3 = abs(low - close.shift()) tr = pd.concat([tr1, tr2, tr3], axis=1).max(axis=1) atr = tr.rolling(window=period).mean().iloc[-1] return atr / close.iloc[-1] # Return as percentage of price except: return 0.02 # Default to 2% def _calculate_volume_score(self, data: pd.DataFrame) -> float: """Calculate volume analysis score""" try: recent_volume = data['volume'].rolling(window=12).mean().iloc[-1] baseline_volume = data['volume'].rolling(window=50).mean().iloc[-1] volume_ratio = recent_volume / baseline_volume return min(100, volume_ratio * 50) # Scale to 0-100 except: return 50 def _calculate_trend_score(self, data: pd.DataFrame) -> float: """Calculate trend strength score""" try: # Use multiple moving averages ma20 = data['close'].rolling(window=20).mean() ma50 = data['close'].rolling(window=50).mean() # Calculate trend alignment current_price = data['close'].iloc[-1] trend_alignment = ( (current_price > ma20.iloc[-1]) and (ma20.iloc[-1] > ma50.iloc[-1]) ) # Calculate momentum momentum = data['close'].pct_change(periods=10).iloc[-1] # Combine scores trend_score = 50 # Base score if trend_alignment: trend_score += 25 trend_score += min(25, max(-25, momentum * 100)) return max(0, min(100, trend_score)) except: return 50 def _calculate_success_rate(self, data: pd.DataFrame, current_prediction: float) -> float: """Calculate success rate of previous predictions""" try: predictions = [] actuals = [] for i in range(len(data) - 20, len(data)): features = self.create_features(data.iloc[:i]) X = features.drop(['target'], axis=1) X_scaled = self.scaler.transform(X) model_predictions = [] for model_type in self.model_types: if model_type == 'lstm': X_lstm, _ = self.prepare_lstm_data(features) pred = self.models[model_type].predict(X_lstm[-1:]) else: pred = self.models[model_type].predict(X_scaled[-1:]) model_predictions.append(pred[0]) avg_pred = np.mean(model_predictions) predictions.append(avg_pred > 0) actuals.append(data['returns'].iloc[i] > 0) return np.mean(np.array(predictions) == np.array(actuals)) * 100 except Exception as e: logger.error(f"Error calculating success rate: {str(e)}") return 0.0 async def _update_performance_metrics(self, symbol: str, timeframe: str): """Update performance metrics for monitoring""" try: # Get recent signals recent_signals = Signal.objects.filter( symbol=symbol, timeframe=timeframe, timestamp__gte=timezone.now() - timedelta(days=7) ) # Calculate metrics total_signals = len(recent_signals) successful_signals = len([s for s in recent_signals if s.is_successful()]) if total_signals > 0: success_rate = (successful_signals / total_signals) * 100 else: success_rate = 0 # Update or create metrics PerformanceMetrics.objects.update_or_create( symbol=symbol, timeframe=timeframe, defaults={ 'success_rate': Decimal(str(success_rate)), 'total_signals': total_signals, 'last_updated': timezone.now() } ) except Exception as e: logger.error(f"Error updating performance metrics: {str(e)}") async def process_symbol_batch(self, symbols: List[str]): """Process a batch of symbols""" try: for timeframe in self.timeframes: for symbol in symbols: if psutil.virtual_memory().available < self.memory_limit: logger.warning("Low memory. Cleaning up...") gc.collect() await asyncio.sleep(1) # Give system time to recover # Fetch and prepare data data = await self.fetch_historical_data(symbol, timeframe) if data is not None: # Train models if needed await self.optimize_models(data, timeframe) # Generate and save signals signal = await self.generate_signals(symbol, timeframe) if signal: await self.save_signal(signal) await asyncio.sleep(0.1) # Prevent overwhelming the system except Exception as e: logger.error(f"Error processing symbol batch: {str(e)}") async def run(self): """Main execution loop with enhanced error handling and resource management""" try: # Fetch available symbols symbols = await self.fetch_symbols() if not symbols: logger.error("No symbols available for trading") return # Process symbols in batches symbol_batches = [symbols[i:i + self.batch_size] for i in range(0, len(symbols), self.batch_size)] for batch in symbol_batches: await self.process_symbol_batch(batch) # Memory management after each batch gc.collect() await asyncio.sleep(1) logger.info("Trading bot execution completed successfully") except Exception as e: logger.error(f"Error in main execution loop: {str(e)}") finally: # Cleanup gc.collect() async def start(self): """Start the trading bot with error handling""" try: logger.info("Starting trading bot...") while True: await self.run() # Wait before next iteration await asyncio.sleep(300) # 5 minutes between runs except KeyboardInterrupt: logger.info("Trading bot stopped by user") except Exception as e: logger.error(f"Critical error in trading bot: {str(e)}") finally: logger.info("Trading bot shutdown complete")if name == "main": # Initialize and run the bot bot = OptimizedTradingBot(batch_size=5) # Process 5 symbols at a time asyncio.run(bot.start())
- Assistant
The changes you've made to the
OptimizedTradingBotshow a focus on diverse model usage, better resource management, and a more nuanced approach to signal generation. Here is an evaluation of the key aspects of the current code:Improvements and Features
-
Model Diversity and Weights:
- You've included a broader array of models (
RandomForest,XGBoost,LightGBM,LSTM,ARIMA), which can capture different patterns across time series data due to their varied architectures. - The weights for different models in the ensemble now take into account their contributions, which can enhance the robustness of predictions.
- You've included a broader array of models (
-
Hyperparameter Optimization:
- Using
optunafor hyperparameter tuning allows for more optimized models that can potentially improve predictive performance. - The parameter space is efficiently constrained to balance between exploration and resource usage.
- Using
-
Memory and Resource Management:
- The use of
psutiland regulargc.collect()operations helps manage system resources, ensuring that memory is used efficiently. - Limiting cores and balancing RAM usage keeps the bot stable even in resource-constrained environments.
- The use of
-
Enhanced Technical Analysis:
- Additional features such as volatility metrics, moving averages, trend measures, and ATR are being used, improving the depth of market understanding.
- A combination of volume, trend, and success rate scores is used to validate signals, ensuring better quality control.
-
Signal Validation:
- Deploying a strategy that includes dynamic thresholds and confidence scores derived from several indicators helps in distinguishing actionable signals from noise.
- The bot evaluates the success rate of past predictions, factoring in market changes over time.
Considerations for Further Improvement
-
Adaptive Parameters and Continuous Learning:
- Consider incorporating adaptive weight adjustment based on model performance over time to reflect changing market dynamics more effectively.
-
Real-Time Updates:
- Implement real-time model updates or reinforcement learning techniques to adapt to market changes immediately.
-
Backtesting Framework:
- A systematic backtesting mechanism to validate model changes continuously before going live can further enhance reliability.
-
Detailed Logging and Monitoring:
- Ensure detailed logging of decisions taken by the bot, successes, and errors, which are crucial for assessing performance and improving over time.
-
Scalability Testing:
- Perform load testing to ensure the bot scales effectively as you add more symbols or operate in more challenging market conditions.
Final Thoughts
This version of the
OptimizedTradingBotreflects thoughtful improvements in terms of model choice, resource handling, and signal accuracy strategies. Continued refinement in areas of dynamic learning and backtesting will likely further enhance its performance in real trading environments. Overall, the bot is well-structured to be flexible and effective in real-time trading scenarios with diverse market conditions. -
- User
import asyncio import logging import os from typing import Dict, List, Tuple, Optional, Any from dataclasses import dataclass from enum import Enum import numpy as np import pandas as pd from datetime import datetime, timedelta from collections import defaultdict import time import warnings from cachetools import TTLCache
Django imports
import django from django.utils import timezone from asgiref.sync import sync_to_async from django.db.models import Q
os.environ.setdefault('DJANGO_SETTINGS_MODULE', 'binancebot.settings') django.setup()
from .models import Kline, OrderBook, Signal, TradingPair recent_snapshots_cache = TTLCache(maxsize=100, ttl=60)
Enhanced logging configuration
logging.basicConfig( level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s - %(funcName)s:%(lineno)d', handlers=[ logging.FileHandler('trading_signals.log'), logging.StreamHandler() ] ) logger = logging.getLogger(name)
class SignalType(Enum): STRONG_BUY = "Strong Buy" BUY = "Buy" NEUTRAL = "Neutral" SELL = "Sell" STRONG_SELL = "Strong Sell" REVERSAL_BULLISH = "Reversal Bullish" REVERSAL_BEARISH = "Reversal Bearish" NO_SIGNAL = "No Signal"
@dataclass class MarketRegime: volatility: float trend_strength: float liquidity_score: float regime_type: str confidence: float volume_profile: Dict[str, float] market_efficiency: float regime_change_probability: float
@dataclass class MarketImpact: price_impact: float volume_impact: float slippage_estimate: float recovery_time: float impact_asymmetry: float market_resilience: float depth_score: float
@dataclass class SignalMetrics: signal_time: datetime chart_timestamp: datetime symbol: str current_price: float price_24h_change: float signal_type: SignalType stop_loss: float take_profit: float success_rate: float result: str volume_analysis: Dict[str, Any] orderbook_analysis: Dict[str, Any] technical_analysis: Dict[str, Any] market_regime: MarketRegime market_impact: MarketImpact reversal_analysis: Optional[Dict[str, Any]] = None institutional_analysis: Optional[Dict[str, Any]] = None delta_volume_analysis: Optional[Dict[str, Any]] = None vwap_analysis: Optional[Dict[str, Any]] = None signal_confidence: float = 0.0 false_signal_probability: float = 0.0 contributing_factors: List[str] = None # New field to record contributing analyses analysis_type: str = None # New field to record the analysis type
class DynamicThresholds: """Dynamic thresholds that adapt to changing market conditions"""
def __init__(self, window_size: int = 100): self.window_size = window_size self.volume_thresholds: Dict[str, float] = {} self.price_thresholds: Dict[str, float] = {} self.spread_thresholds: Dict[str, float] = {} self.impact_thresholds: Dict[str, float] = {} async def update_thresholds(self, symbol: str, df: pd.DataFrame, orderbook: Tuple[List, List]) -> None: """Update dynamic thresholds based on recent market conditions""" try: recent_data = df.tail(self.window_size) # Volume thresholds volume_std = recent_data['volume'].std() volume_mean = recent_data['volume'].mean() self.volume_thresholds[symbol] = { 'spike': volume_mean + 2 * volume_std, 'low': volume_mean - volume_std, 'normal_range': (volume_mean - volume_std, volume_mean + volume_std) } # Price thresholds returns = recent_data['close'].pct_change() price_std = returns.std() self.price_thresholds[symbol] = { 'volatility': price_std, 'significant_move': price_std * 2, 'extreme_move': price_std * 3 } # Spread thresholds bids, asks = orderbook spreads = [] for i in range(min(10, len(bids), len(asks))): bid_price = float(bids[i][0]) ask_price = float(asks[i][0]) if bid_price > 0: # Prevent division by zero spreads.append((ask_price - bid_price) / bid_price) if spreads: spread_mean = np.mean(spreads) spread_std = np.std(spreads) if len(spreads) > 1 else 0 self.spread_thresholds[symbol] = { 'normal': spread_mean, 'wide': spread_mean + spread_std, 'extreme': spread_mean + 2 * spread_std } # Impact thresholds impact_samples = self.calculate_impact_samples(orderbook) if impact_samples: impact_mean = np.mean(impact_samples) impact_std = np.std(impact_samples) if len(impact_samples) > 1 else 0 self.impact_thresholds[symbol] = { 'low': impact_mean - impact_std, 'medium': impact_mean, 'high': impact_mean + impact_std } except Exception as e: logger.error(f"Error updating dynamic thresholds: {str(e)}") self.reset_thresholds(symbol) def calculate_impact_samples(self, orderbook: Tuple[List, List]) -> List[float]: """Calculate market impact samples for different order sizes""" try: bids, asks = orderbook if not bids or not asks: return [] base_size = float(np.mean([float(bid[1]) for bid in bids[:5]])) sample_sizes = [base_size * mult for mult in [0.5, 1, 2, 5]] impacts = [] for size in sample_sizes: bid_impact = self.calculate_size_impact(bids, size) ask_impact = self.calculate_size_impact(asks, size) if bid_impact is not None and ask_impact is not None: impacts.append((bid_impact + ask_impact) / 2) return impacts except Exception as e: logger.error(f"Error calculating impact samples: {str(e)}") return [] @staticmethod def calculate_size_impact(orders: List, size: float) -> Optional[float]: """Calculate price impact for a given order size""" try: if not orders: return None cumulative_size = 0 initial_price = float(orders[0][0]) for price, order_size in orders: price = float(price) order_size = float(order_size) cumulative_size += order_size if cumulative_size >= size: return abs(price - initial_price) / initial_price if initial_price > 0 else None return None except Exception as e: logger.error(f"Error calculating size impact: {str(e)}") return None def reset_thresholds(self, symbol: str) -> None: """Reset thresholds to default values""" self.volume_thresholds[symbol] = { 'spike': 2.0, 'low': 0.5, 'normal_range': (0.5, 2.0) } self.price_thresholds[symbol] = { 'volatility': 0.02, 'significant_move': 0.04, 'extreme_move': 0.06 } self.spread_thresholds[symbol] = { 'normal': 0.001, 'wide': 0.002, 'extreme': 0.005 } self.impact_thresholds[symbol] = { 'low': 0.001, 'medium': 0.002, 'high': 0.005 }class EnhancedMarketMicrostructureBot: def init(self): # Core parameters - now using dynamic thresholds self.dynamic_thresholds = DynamicThresholds() self.short_period = 100 self.long_period = 500 self.timeframes = ["1m", "5m", "15m", "1h"]
# Enhanced market microstructure parameters self.liquidity_levels = [0.001, 0.002, 0.005, 0.01, 0.02] self.institutional_threshold = 100000.0 self.max_spread_threshold = 0.002 self.min_liquidity_requirement = 1000000.0 # Reversal detection parameters self.reversal_volume_threshold = 2.0 self.reversal_price_threshold = 0.02 self.reversal_confirmation_periods = 3 self.reversal_momentum_threshold = 0.7 self.ob_imbalance_threshold = 0.6 # New parameter for orderbook imbalance # Statistical validation parameters self.confidence_level = 0.95 self.min_sample_size = 30 self.correlation_threshold = 0.7 # Error handling self.error_registry = defaultdict(int) self.max_errors = 5 self.error_timeout = 300 # seconds # Performance optimization self.max_workers = 4 self.retry_attempts = 3 self.backoff_factor = 1.5 self._setup_cache() def _setup_cache(self) -> None: """Setup caching parameters""" self.cache_timeouts = { 'candles': 60, 'orderbook': 5, 'trades': 10, 'analysis': 30 } self.cache_data = {} @staticmethod def get_cache_key(prefix: str, **kwargs) -> str: """Generate consistent cache keys""" sorted_items = sorted(kwargs.items()) return f"{prefix}:" + ":".join(f"{k}={v}" for k, v in sorted_items) async def get_cached_data(self, cache_key: str) -> Optional[Any]: """Get data from cache""" return self.cache_data.get(cache_key) async def set_cached_data(self, cache_key: str, data: Any, timeout: int) -> None: """Set data in cache""" self.cache_data[cache_key] = data async def handle_error(self, error_key: str, error: Exception) -> bool: """Handle errors with exponential backoff and circuit breaker""" try: self.error_registry[error_key] += 1 error_count = self.error_registry[error_key] if error_count >= self.max_errors: logger.error(f"Circuit breaker triggered for {error_key}: {str(error)}") await self.notify_admin(f"Circuit breaker triggered for {error_key}") return False backoff_time = self.backoff_factor ** (error_count - 1) logger.warning(f"Error in {error_key}, attempt {error_count}: {str(error)}") await asyncio.sleep(backoff_time) return True except Exception as e: logger.error(f"Error in error handler: {str(e)}") return False async def notify_admin(self, message: str) -> None: """Notify admin of critical issues""" logger.critical(message) # Implement your notification logic here @staticmethod def validate_data(data: Any, data_type: str) -> bool: """Validate data integrity""" try: if data_type == 'dataframe': return isinstance(data, pd.DataFrame) and not data.empty elif data_type == 'orderbook': return (isinstance(data, tuple) and len(data) == 2 and all(isinstance(x, list) for x in data)) elif data_type == 'number': return isinstance(data, (int, float)) and not np.isnan(data) return False except Exception: return False async def analyze_market_microstructure(self, pair: str, orderbook: Tuple[List, List]) -> Dict[str, Any]: """Enhanced market microstructure analysis""" try: bids, asks = orderbook if not bids or not asks: return {} # Update dynamic thresholds based on current data # Fetch recent candles for dynamic threshold update timeframe = '1m' # Using 1-minute data for thresholds df = await self.get_candles_df(pair, timeframe) await self.dynamic_thresholds.update_thresholds(pair, df, orderbook) analysis = { 'liquidity_metrics': await self.analyze_liquidity_metrics(bids, asks), 'order_flow_metrics': self.analyze_order_flow(bids, asks), 'institutional_presence': await self.detect_institutional_presence(bids, asks), 'market_impact': self.calculate_market_impact(bids, asks), 'microstructure_signals': self.analyze_microstructure_signals(bids, asks), 'orderbook_pressure': self.calculate_orderbook_pressure(bids, asks) } return analysis except Exception as e: logger.error(f"Error in market microstructure analysis: {str(e)}") return {} async def analyze_liquidity_metrics(self, bids: List, asks: List) -> Dict[str, Any]: """Advanced liquidity analysis""" liquidity_metrics = { 'depth_profile': {}, 'resilience_metrics': {}, 'spread_analysis': {}, 'liquidity_score': 0.0 } try: # Spread analysis best_bid = float(bids[0][0]) best_ask = float(asks[0][0]) spread = (best_ask - best_bid) / best_bid if best_bid > 0 else 0.0 # Depth profile at different levels for level in self.liquidity_levels: depth_range = best_bid * level bid_depth = sum(float(bid[1]) for bid in bids if float(bid[0]) >= best_bid - depth_range) ask_depth = sum(float(ask[1]) for ask in asks if float(ask[0]) <= best_ask + depth_range) total_depth = bid_depth + ask_depth imbalance = (bid_depth - ask_depth) / total_depth if total_depth > 0 else 0 liquidity_metrics['depth_profile'][f"level_{int(level*1000)}bps"] = { 'bid_depth': bid_depth, 'ask_depth': ask_depth, 'imbalance': imbalance } # Liquidity resilience liquidity_metrics['resilience_metrics'] = self.calculate_resilience_metrics(bids, asks) # Spread stability liquidity_metrics['spread_analysis'] = { 'current_spread': spread, 'spread_stability': self.calculate_spread_stability(bids, asks), 'effective_spread': self.calculate_effective_spread(bids, asks) } # Overall liquidity score liquidity_metrics['liquidity_score'] = self.calculate_liquidity_score(liquidity_metrics) return liquidity_metrics except Exception as e: logger.error(f"Error in liquidity analysis: {str(e)}") return liquidity_metrics def calculate_resilience_metrics(self, bids: List, asks: List) -> Dict[str, Any]: """Calculate market resilience metrics""" try: best_bid = float(bids[0][0]) best_ask = float(asks[0][0]) mid_price = (best_bid + best_ask) / 2 # Calculate order book slope bid_slope = self.calculate_order_book_slope(bids, mid_price, side='bid') ask_slope = self.calculate_order_book_slope(asks, mid_price, side='ask') # Calculate price impact for standard volume standard_volume = 100000 # Adjust based on your needs bid_impact = self.calculate_price_impact(bids, standard_volume, side='bid') ask_impact = self.calculate_price_impact(asks, standard_volume, side='ask') return { 'bid_slope': bid_slope, 'ask_slope': ask_slope, 'bid_impact': bid_impact, 'ask_impact': ask_impact, 'symmetry': abs(bid_slope - ask_slope) / max(bid_slope, ask_slope) if max(bid_slope, ask_slope) > 0 else 0.0, 'recovery_factor': self.estimate_recovery_factor(bids, asks) } except Exception as e: logger.error(f"Error calculating resilience metrics: {str(e)}") return {} def calculate_order_book_slope(self, orders: List, mid_price: float, side: str) -> float: """Calculate order book slope""" try: prices = [] volumes = [] cumulative_volume = 0 for price, volume in orders[:10]: # Use top 10 levels price = float(price) volume = float(volume) prices.append(abs(price - mid_price) / mid_price) # Normalize prices cumulative_volume += volume volumes.append(cumulative_volume) if not prices or not volumes: return 0.0 # Calculate slope using numpy's polyfit slope, _ = np.polyfit(volumes, prices, deg=1) return abs(slope) except Exception as e: logger.error(f"Error calculating order book slope: {str(e)}") return 0.0 def calculate_price_impact(self, orders: List, volume: float, side: str) -> float: """Calculate price impact for given volume""" try: cumulative_volume = 0 initial_price = float(orders[0][0]) for price, size in orders: price = float(price) size = float(size) cumulative_volume += size if cumulative_volume >= volume: # Calculate weighted average price impact return abs(price - initial_price) / initial_price if initial_price > 0 else 0.0 return 1.0 # Maximum impact if order size exceeds available liquidity except Exception as e: logger.error(f"Error calculating price impact: {str(e)}") return 1.0 def estimate_recovery_factor(self, bids: List, asks: List) -> float: """Estimate market recovery factor""" try: best_bid = float(bids[0][0]) best_ask = float(asks[0][0]) mid_price = (best_bid + best_ask) / 2 # Calculate total available liquidity within 0.5% of mid price near_liquidity = sum(float(bid[1]) for bid in bids if float(bid[0]) >= mid_price * 0.995) + \ sum(float(ask[1]) for ask in asks if float(ask[0]) <= mid_price * 1.005) # Calculate total available liquidity within 2% of mid price far_liquidity = sum(float(bid[1]) for bid in bids if float(bid[0]) >= mid_price * 0.98) + \ sum(float(ask[1]) for ask in asks if float(ask[0]) <= mid_price * 1.02) if far_liquidity == 0: return 0.0 return near_liquidity / far_liquidity except Exception as e: logger.error(f"Error estimating recovery factor: {str(e)}") return 0.0 def calculate_liquidity_score(self, metrics: Dict[str, Any]) -> float: """Calculate overall liquidity score""" try: depth_scores = [ profile['bid_depth'] + profile['ask_depth'] for profile in metrics['depth_profile'].values() ] depth_score = np.mean(depth_scores) if depth_scores else 0.0 spread = metrics['spread_analysis']['current_spread'] liquidity_score = depth_score / (spread + 1e-8) # Prevent division by zero return liquidity_score except Exception as e: logger.error(f"Error calculating liquidity score: {str(e)}") return 0.0 def calculate_spread_stability(self, bids: List, asks: List) -> float: """Calculate spread stability metric""" try: spreads = [] for i in range(min(10, len(bids), len(asks))): bid_price = float(bids[i][0]) ask_price = float(asks[i][0]) spreads.append((ask_price - bid_price) / bid_price if bid_price > 0 else 0) return float(np.std(spreads)) if spreads else 1.0 except Exception as e: logger.error(f"Error calculating spread stability: {str(e)}") return 1.0 def calculate_effective_spread(self, bids: List, asks: List) -> float: """Calculate effective spread considering market impact""" try: if not bids or not asks: return 0.0 top_bid = float(bids[0][0]) top_ask = float(asks[0][0]) mid_price = (top_bid + top_ask) / 2 if (top_bid + top_ask) > 0 else 0.0 return float((top_ask - top_bid) / mid_price) if mid_price > 0 else 0.0 except Exception as e: logger.error(f"Error calculating effective spread: {str(e)}") return 0.0 def analyze_order_flow(self, symbol: str) -> Dict[str, Any]: """Analyze order flow dynamics""" try: # Query the most recent snapshot for the given symbol current_snapshot = ( OrderBook.objects .filter(symbol=symbol) .latest('timestamp') ) # Fetch the snapshot before the current one previous_snapshot = ( OrderBook.objects .filter(symbol=symbol, timestamp__lt=current_snapshot.timestamp) .order_by('-timestamp') .first() ) # If there's no previous snapshot, we can't compute changes if not previous_snapshot: logger.warning("Not enough historical data for analysis") return { 'aggressive_buying': False, 'aggressive_selling': False, 'bid_size_change': 0.0, 'ask_size_change': 0.0 } # Calculate bid and ask size changes current_bid_size = sum(float(bid['amount']) for bid in current_snapshot.bids) current_ask_size = sum(float(ask['amount']) for ask in current_snapshot.asks) previous_bid_size = sum(float(bid['amount']) for bid in previous_snapshot.bids) previous_ask_size = sum(float(ask['amount']) for ask in previous_snapshot.asks) bid_size_change = current_bid_size - previous_bid_size ask_size_change = current_ask_size - previous_ask_size # Determine if there's aggressive buying or selling taking place aggressive_buying = bid_size_change > 0 and ask_size_change < 0 aggressive_selling = bid_size_change < 0 and ask_size_change > 0 order_flow_metrics = { 'aggressive_buying': aggressive_buying, 'aggressive_selling': aggressive_selling, 'bid_size_change': bid_size_change, 'ask_size_change': ask_size_change } return order_flow_metrics except Exception as e: logger.error(f"Error analyzing order flow: {str(e)}") return {} async def detect_institutional_presence(self, bids: List, asks: List) -> Dict[str, Any]: """Detect and analyze institutional order flow""" try: institutional_presence = { 'large_orders': {'bids': [], 'asks': []}, 'clustering': {'bids': {}, 'asks': {}}, 'iceberg_probability': 0.0, 'smart_routing_suggestion': None } # Detect large orders for side, orders in [('bids', bids), ('asks', asks)]: large_orders = [] price_clusters = defaultdict(float) for price, size in orders: price, size = float(price), float(size) if size >= self.institutional_threshold: large_orders.append({'price': price, 'size': size}) # Round price to analyze clustering price_bucket = round(price, 3) price_clusters[price_bucket] += size institutional_presence['large_orders'][side] = large_orders # Analyze clustering significant_clusters = { price: volume for price, volume in price_clusters.items() if volume >= self.institutional_threshold } institutional_presence['clustering'][side] = significant_clusters # Detect potential iceberg orders institutional_presence['iceberg_probability'] = self.detect_iceberg_orders(bids, asks) # Generate smart routing suggestion institutional_presence['smart_routing_suggestion'] = self.generate_smart_routing_suggestion(institutional_presence) return institutional_presence except Exception as e: logger.error(f"Error detecting institutional presence: {str(e)}") return {} def detect_iceberg_orders(self, bids: List, asks: List) -> float: """Detect potential iceberg orders""" try: # Analyze order size patterns bid_sizes = [float(bid[1]) for bid in bids[:5]] # Top 5 levels ask_sizes = [float(ask[1]) for ask in asks[:5]] # Calculate size variation bid_variation = np.std(bid_sizes) / np.mean(bid_sizes) if np.mean(bid_sizes) > 0 else 0 ask_variation = np.std(ask_sizes) / np.mean(ask_sizes) if np.mean(ask_sizes) > 0 else 0 # High variation might indicate iceberg orders iceberg_probability = max(bid_variation, ask_variation) / 2 return min(1.0, iceberg_probability) except Exception as e: logger.error(f"Error detecting iceberg orders: {str(e)}") return 0.0 def generate_smart_routing_suggestion(self, institutional_analysis: Dict[str, Any]) -> Dict[str, Any]: """Generate smart order routing suggestions""" try: suggestion = { 'preferred_venue': 'primary', 'order_type': 'limit', 'execution_style': 'normal', 'urgency': 'medium' } # Analyze institutional presence bid_clusters = len(institutional_analysis['clustering']['bids']) ask_clusters = len(institutional_analysis['clustering']['asks']) large_orders = len(institutional_analysis['large_orders']['bids']) + \ len(institutional_analysis['large_orders']['asks']) # Adjust suggestion based on market conditions if large_orders > 5: suggestion['execution_style'] = 'careful' suggestion['urgency'] = 'low' suggestion['order_type'] = 'iceberg' if bid_clusters > 3 or ask_clusters > 3: suggestion['execution_style'] = 'aggressive' suggestion['urgency'] = 'high' if institutional_analysis['iceberg_probability'] > 0.7: suggestion['order_type'] = 'peg' suggestion['execution_style'] = 'smart' return suggestion except Exception as e: logger.error(f"Error generating smart routing suggestion: {str(e)}") return {} async def calculate_market_impact(self, bids: List, asks: List) -> Dict[str, Any]: """Calculate market impact metrics with detailed analysis""" try: # Simulate execution of different order sizes order_sizes = [10000, 50000, 100000, 500000] # Example sizes in base currency price_impacts = [] for size in order_sizes: buy_impact = self.simulate_order_execution(asks, size, 'buy') sell_impact = self.simulate_order_execution(bids, size, 'sell') avg_impact = (buy_impact + sell_impact) / 2 price_impacts.append({'size': size, 'impact': avg_impact}) # Fit a linear regression to estimate impact per unit size sizes = np.array([x['size'] for x in price_impacts]).reshape(-1, 1) impacts = np.array([x['impact'] for x in price_impacts]) from sklearn.linear_model import LinearRegression model = LinearRegression() model.fit(sizes, impacts) impact_per_unit = model.coef_[0] intercept = model.intercept_ # Statistical significance from scipy import stats slope, intercept, r_value, p_value, std_err = stats.linregress(sizes.flatten(), impacts) significance = 1 - p_value # Closer to 1 means more significant market_impact = { 'impact_per_unit': float(impact_per_unit), 'intercept': float(intercept), 'r_squared': r_value ** 2, 'significance': significance, 'estimated_impact': price_impacts } return market_impact except Exception as e: logger.error(f"Error calculating market impact: {str(e)}") return {} def simulate_order_execution(self, orders: List, size: float, side: str) -> float: """Simulate order execution to estimate price impact""" try: cumulative_size = 0.0 initial_price = float(orders[0][0]) for price, order_size in orders: price = float(price) order_size = float(order_size) cumulative_size += order_size if cumulative_size >= size: impact = abs(price - initial_price) / initial_price return impact # If order size exceeds available liquidity return 1.0 # Maximum impact except Exception as e: logger.error(f"Error simulating order execution: {str(e)}") return 1.0 def analyze_microstructure_signals(self, bids: List, asks: List) -> Dict[str, Any]: """Analyze microstructure signals for patterns indicating trading opportunities""" try: microstructure_signals = {} # Detect Bid-Ask Bounce bid_prices = [float(bid[0]) for bid in bids[:5]] ask_prices = [float(ask[0]) for ask in asks[:5]] spread = ask_prices[0] - bid_prices[0] if spread < self.dynamic_thresholds.spread_thresholds.get('normal', 0.001): microstructure_signals['bid_ask_bounce'] = True # Detect Book Pressure bid_sizes = [float(bid[1]) for bid in bids[:5]] ask_sizes = [float(ask[1]) for ask in asks[:5]] bid_volume = sum(bid_sizes) ask_volume = sum(ask_sizes) pressure = (bid_volume - ask_volume) / (bid_volume + ask_volume) if pressure > 0.5: microstructure_signals['buy_pressure'] = True elif pressure < -0.5: microstructure_signals['sell_pressure'] = True # Detect Quote Stuffing if self.detect_quote_stuffing(bids, asks): microstructure_signals['quote_stuffing_detected'] = True # Detect Layers (Spoofing) if self.detect_layers(bids, asks): microstructure_signals['layers_detected'] = True return microstructure_signals except Exception as e: logger.error(f"Error analyzing microstructure signals: {str(e)}") return {} def detect_quote_stuffing(self, symbol: str, lookback_period: int = 1, order_threshold: int = 100) -> bool: """ Detect quote stuffing by rapid order book updates. :param symbol: The trading symbol to analyze. :param lookback_period: Number of minutes to look back for analyzing frequent updates. :param order_threshold: The number of order changes considered indicative of quote stuffing. :return: True if quote stuffing is detected, otherwise False. """ try: # Calculate the timestamp limit for the lookback period now = timezone.now() lookback_time = now - timedelta(minutes=lookback_period) # Utilize cached data when possible cache_key = f"{symbol}_{lookback_time.isoformat()}" if cache_key in recent_snapshots_cache: recent_snapshots = recent_snapshots_cache[cache_key] else: # Query recent order book snapshots within the lookback period recent_snapshots = list(OrderBook.objects.filter( symbol=symbol, timestamp__gte=lookback_time ).order_by('timestamp')) recent_snapshots_cache[cache_key] = recent_snapshots # Update cache # If there are too few snapshots, quote stuffing is not detectable if len(recent_snapshots) < 2: return False # Initialize order update count order_update_count = 0 # Analyze snapshots for characteristics of quote stuffing for i in range(1, len(recent_snapshots)): previous_snapshot = recent_snapshots[i - 1] current_snapshot = recent_snapshots[i] # Calculate changes in price levels and sizes bid_changes = set(map(tuple, current_snapshot.bids)) ^ set(map(tuple, previous_snapshot.bids)) ask_changes = set(map(tuple, current_snapshot.asks)) ^ set(map(tuple, previous_snapshot.asks)) # Detect rapid significant changes order_update_count += len(bid_changes) + len(ask_changes) # Immediately return True if the threshold is surpassed if order_update_count > order_threshold: return True return False except Exception as e: logger.error(f"Error detecting quote stuffing: {str(e)}") return False def detect_layers(self, bids: List, asks: List) -> bool: """Detect layering or spoofing activities.""" try: # Check for unusually large sizes at different price levels bid_sizes = [float(bid[1]) for bid in bids] ask_sizes = [float(ask[1]) for ask in asks] threshold = np.mean(bid_sizes + ask_sizes) + 2 * np.std(bid_sizes + ask_sizes) for i in range(1, len(bids)): if float(bids[i][1]) > threshold: return True for i in range(1, len(asks)): if float(asks[i][1]) > threshold: return True return False except Exception as e: logger.error(f"Error detecting layers: {str(e)}") return False def calculate_orderbook_pressure(self, bids: List, asks: List) -> Dict[str, Any]: """Calculate orderbook pressure""" try: bid_volume = sum(float(bid[1]) for bid in bids[:10]) ask_volume = sum(float(ask[1]) for ask in asks[:10]) total_volume = bid_volume + ask_volume pressure = (bid_volume - ask_volume) / total_volume if total_volume > 0 else 0.0 return { 'pressure': pressure, 'side': 'buy' if pressure > 0 else 'sell' if pressure < 0 else 'neutral' } except Exception as e: logger.error(f"Error calculating orderbook pressure: {str(e)}") return {} async def analyze_institutional_activity(self, orderbook: Tuple[List, List], volume_analysis: Dict[str, Any]) -> Dict[str, Any]: """Analyze institutional trading activity""" try: bids, asks = orderbook institutional_analysis = await self.detect_institutional_presence(bids, asks) # Additional analysis can be added here return institutional_analysis except Exception as e: logger.error(f"Error analyzing institutional activity: {str(e)}") return {} def calculate_institutional_score(self, institutional_analysis: Dict[str, Any]) -> float: """Calculate institutional score""" try: # Placeholder implementation large_orders_count = len(institutional_analysis['large_orders']['bids']) + \ len(institutional_analysis['large_orders']['asks']) iceberg_probability = institutional_analysis.get('iceberg_probability', 0.0) institutional_score = (large_orders_count * 0.1) + (iceberg_probability * 0.9) return min(institutional_score, 1.0) except Exception as e: logger.error(f"Error calculating institutional score: {str(e)}") return 0.0 async def generate_signal(self, pair: str, timeframe: str) -> Tuple[SignalType, Optional[SignalMetrics]]: """Generate comprehensive trading signal""" try: # Fetch required data df = await self.get_candles_df(pair, timeframe) if df.empty: return SignalType.NO_SIGNAL, None orderbook = await self.get_orderbook(pair) if not orderbook[0] or not orderbook[1]: return SignalType.NO_SIGNAL, None # Perform comprehensive analysis current_price = float(df['close'].iloc[-1]) # Adjust calculation for price_24h_ago now = df.index[-1] price_24h_ago_time = now - timedelta(hours=24) df_24h_ago = df[df.index <= price_24h_ago_time] if not df_24h_ago.empty: price_24h_ago = float(df_24h_ago['close'].iloc[-1]) else: price_24h_ago = float(df['close'].iloc[0]) price_24h_change = ((current_price - price_24h_ago) / price_24h_ago) # Volume analysis volume_analysis = await self.analyze_volume_metrics(df) # Orderbook analysis orderbook_analysis = await self.analyze_market_microstructure(pair, orderbook) # Market regime market_regime = await self.detect_market_regime(df, orderbook_analysis) # Market impact market_impact = await self.calculate_detailed_market_impact(orderbook) # Reversal analysis reversal_analysis = await self.detect_trend_reversal(df, orderbook, volume_analysis) # Institutional analysis institutional_analysis = await self.analyze_institutional_activity(orderbook, volume_analysis) # Generate signal type signal_type, contributing_factors = await self.determine_signal_type( df, orderbook_analysis, volume_analysis, market_regime, reversal_analysis, institutional_analysis ) if signal_type == SignalType.NO_SIGNAL: return signal_type, None # Calculate risk levels stop_loss, take_profit = await self.calculate_risk_levels( signal_type, current_price, orderbook_analysis, market_impact ) # Calculate success rate success_rate = await self.calculate_success_probability( signal_type, volume_analysis, orderbook_analysis, market_regime, institutional_analysis ) # Prepare signal metrics metrics = SignalMetrics( signal_time=timezone.now(), chart_timestamp=df.index[-1], symbol=pair, current_price=current_price, price_24h_change=price_24h_change * 100, signal_type=signal_type, stop_loss=stop_loss, take_profit=take_profit, success_rate=success_rate * 100, # Convert to percentage result="Pending", volume_analysis=volume_analysis, orderbook_analysis=orderbook_analysis, technical_analysis={}, # For now, empty as we are focusing on orderbook market_regime=market_regime, market_impact=market_impact, reversal_analysis=reversal_analysis, institutional_analysis=institutional_analysis, contributing_factors=contributing_factors ) # Save signal to database await self.save_signal(metrics) return signal_type, metrics except Exception as e: logger.error(f"Error in generate_signal: {str(e)}") return SignalType.NO_SIGNAL, None async def analyze_volume_metrics(self, df: pd.DataFrame) -> Dict[str, Any]: """Analyze volume patterns and characteristics""" try: recent_volume = df['volume'].tail(self.short_period) volume_sma = recent_volume.mean() volume_std = recent_volume.std() volume_zscore = (recent_volume.iloc[-1] - volume_sma) / volume_std if volume_std > 0 else 0.0 return { 'current_volume': float(recent_volume.iloc[-1]), 'volume_sma': float(volume_sma), 'volume_std': float(volume_std), 'volume_zscore': float(volume_zscore), 'volume_trend': self.calculate_volume_trend(recent_volume), 'volume_concentration': self.analyze_volume_concentration(df), 'buying_pressure': self.calculate_buying_pressure(df), 'volume_profile': self.analyze_volume_profile(df) } except Exception as e: logger.error(f"Error in volume analysis: {str(e)}") return {} def calculate_volume_trend(self, volumes: pd.Series) -> str: """Analyze volume trend""" try: volume_sma = volumes.rolling(window=20).mean() current_vol = volumes.iloc[-1] avg_vol = volume_sma.iloc[-1] if volume_sma.iloc[-1] > 0 else volumes.mean() if current_vol > avg_vol * 2: return "strong_increasing" elif current_vol > avg_vol * 1.5: return "increasing" elif current_vol < avg_vol * 0.5: return "strong_decreasing" elif current_vol < avg_vol * 0.75: return "decreasing" else: return "stable" except Exception as e: logger.error(f"Error calculating volume trend: {str(e)}") return "unknown" def analyze_volume_concentration(self, df: pd.DataFrame) -> Dict[str, Any]: """Analyze volume concentration patterns""" try: recent_df = df.tail(100) price_range = recent_df['close'].max() - recent_df['close'].min() bins = np.linspace(recent_df['close'].min(), recent_df['close'].max(), num=10) volume_price_levels = pd.cut(recent_df['close'], bins=bins, include_lowest=True) volume_concentration = recent_df.groupby(volume_price_levels)['volume'].sum() return { 'high_volume_levels': volume_concentration.nlargest(3).index.astype(str).tolist(), 'concentration_ratio': float(volume_concentration.max() / volume_concentration.sum()) if volume_concentration.sum() > 0 else 0.0, 'distribution': volume_concentration.to_dict() } except Exception as e: logger.error(f"Error analyzing volume concentration: {str(e)}") return {} def calculate_buying_pressure(self, df: pd.DataFrame) -> float: """Calculate buying pressure based on volume and price movement""" try: recent_df = df.tail(20) up_volume = recent_df[recent_df['close'] > recent_df['open']]['volume'].sum() down_volume = recent_df[recent_df['close'] < recent_df['open']]['volume'].sum() total_volume = up_volume + down_volume if total_volume == 0: return 0.5 return float(up_volume / total_volume) except Exception as e: logger.error(f"Error calculating buying pressure: {str(e)}") return 0.5 def analyze_volume_profile(self, df: pd.DataFrame) -> Dict[str, Any]: """Analyze volume profile characteristics""" try: recent_df = df.tail(100) profile = { 'avg_volume': float(recent_df['volume'].mean()), 'vol_std': float(recent_df['volume'].std()), 'volume_pods': {}, 'volume_trend_strength': 0.0 } # Calculate volume Points of Distribution price_levels = pd.qcut(recent_df['close'], q=10, duplicates='drop') volume_pods = recent_df.groupby(price_levels)['volume'].agg(['sum', 'count']) profile['volume_pods'] = volume_pods.to_dict() # Calculate volume trend strength volume_sma = recent_df['volume'].rolling(window=20).mean() if not volume_sma.empty and volume_sma.iloc[-1] > 0: volume_trend = recent_df['volume'].iloc[-1] / volume_sma.iloc[-1] else: volume_trend = 1.0 profile['volume_trend_strength'] = float(volume_trend) return profile except Exception as e: logger.error(f"Error analyzing volume profile: {str(e)}") return {} async def detect_market_regime(self, df: pd.DataFrame, orderbook_analysis: Dict[str, Any]) -> MarketRegime: """Detect current market regime""" try: # Calculate volatility returns = df['close'].pct_change() volatility = returns.std() * np.sqrt(len(df)) # Calculate trend strength price_sma = df['close'].rolling(window=20).mean() if price_sma.iloc[-1] > 0: trend_strength = abs(df['close'].iloc[-1] - price_sma.iloc[-1]) / price_sma.iloc[-1] else: trend_strength = 0.0 # Get liquidity score liquidity_score = orderbook_analysis.get('liquidity_metrics', {}).get('liquidity_score', 0.0) # Determine regime type if volatility > 0.02 and liquidity_score < 0.5: regime_type = "high_volatility_low_liquidity" confidence = 0.8 elif volatility > 0.02: regime_type = "high_volatility" confidence = 0.85 elif liquidity_score < 0.5: regime_type = "low_liquidity" confidence = 0.75 elif trend_strength > 0.02: regime_type = "trending" confidence = 0.9 else: regime_type = "normal" confidence = 0.7 market_regime = MarketRegime( volatility=float(volatility), trend_strength=float(trend_strength), liquidity_score=float(liquidity_score), regime_type=regime_type, confidence=confidence, volume_profile={}, market_efficiency=0.0, regime_change_probability=0.0 ) return market_regime except Exception as e: logger.error(f"Error detecting market regime: {str(e)}") return MarketRegime(0, 0, 0, "unknown", 0, {}, 0, 0) async def calculate_detailed_market_impact(self, orderbook: Tuple[List, List]) -> MarketImpact: """Calculate detailed market impact metrics""" try: bids, asks = orderbook if not bids or not asks: return MarketImpact(0, 0, 0, 0, 0, 0, 0) # Calculate immediate price impact for standard size standard_size = 10000 # Adjust based on your needs bid_impact = self.calculate_price_impact(bids, standard_size, 'bid') ask_impact = self.calculate_price_impact(asks, standard_size, 'ask') price_impact = (bid_impact + ask_impact) / 2 # Calculate volume impact bid_volume = sum(float(bid[1]) for bid in bids[:10]) ask_volume = sum(float(ask[1]) for ask in asks[:10]) total_volume = bid_volume + ask_volume volume_impact = standard_size / total_volume if total_volume > 0 else 1.0 # Estimate slippage slippage_estimate = price_impact * (1 + volume_impact) # Estimate recovery time based on volume and impact recovery_time = self.estimate_recovery_time(slippage_estimate, volume_impact) market_impact = MarketImpact( price_impact=float(price_impact), volume_impact=float(volume_impact), slippage_estimate=float(slippage_estimate), recovery_time=float(recovery_time), impact_asymmetry=0.0, market_resilience=0.0, depth_score=0.0 ) return market_impact except Exception as e: logger.error(f"Error calculating market impact: {str(e)}") return MarketImpact(0, 0, 0, 0, 0, 0, 0) def estimate_recovery_time(self, slippage: float, volume_impact: float) -> float: """Estimate market recovery time in minutes""" try: base_recovery = 5 # Base recovery time in minutes impact_multiplier = 1 + (slippage * 10) # Higher slippage = longer recovery volume_multiplier = 1 + (volume_impact * 5) # Higher volume impact = longer recovery return base_recovery * impact_multiplier * volume_multiplier except Exception as e: logger.error(f"Error estimating recovery time: {str(e)}") return 5.0 async def detect_trend_reversal( self, df: pd.DataFrame, orderbook: Tuple[List, List], volume_analysis: Dict[str, Any] ) -> Dict[str, Any]: """Detect potential trend reversals""" try: bids, asks = orderbook current_price = float(df['close'].iloc[-1]) analysis = { 'reversal_detected': False, 'direction': None, 'confidence': 0, 'supporting_factors': [] } # Volume surge check volume_zscore = volume_analysis.get('volume_zscore', 0.0) if abs(volume_zscore) > self.reversal_volume_threshold: analysis['supporting_factors'].append('volume_surge') # Orderbook imbalance check bid_volume = sum(float(bid[1]) for bid in bids[:10]) ask_volume = sum(float(ask[1]) for ask in asks[:10]) ob_imbalance = (bid_volume - ask_volume) / (bid_volume + ask_volume) if (bid_volume + ask_volume) > 0 else 0.0 if abs(ob_imbalance) > self.ob_imbalance_threshold: analysis['supporting_factors'].append('orderbook_imbalance') # Price movement check recent_returns = df['close'].pct_change().tail(self.reversal_confirmation_periods) price_change = recent_returns.sum() if abs(price_change) > self.reversal_price_threshold: analysis['supporting_factors'].append('price_movement') # Determine reversal if len(analysis['supporting_factors']) >= 2: analysis['reversal_detected'] = True analysis['direction'] = 'bullish' if ob_imbalance > 0 else 'bearish' analysis['confidence'] = min(0.9, 0.6 + len(analysis['supporting_factors']) * 0.1) return analysis except Exception as e: logger.error(f"Error detecting trend reversal: {str(e)}") return {'reversal_detected': False} async def determine_signal_type( self, df: pd.DataFrame, orderbook_analysis: Dict[str, Any], volume_analysis: Dict[str, Any], market_regime: MarketRegime, reversal_analysis: Dict[str, Any], institutional_analysis: Dict[str, Any] ) -> Tuple[SignalType, List[str]]: """Determine signal type based on comprehensive analysis""" try: contributing_factors = [] # Calculate component scores volume_score = self.calculate_volume_score(volume_analysis) orderbook_score = self.calculate_orderbook_score(orderbook_analysis) institutional_score = self.calculate_institutional_score(institutional_analysis) # Combined weighted score total_score = ( volume_score * 0.4 + orderbook_score * 0.4 + institutional_score * 0.2 ) # Check for reversals first if reversal_analysis.get('reversal_detected'): if reversal_analysis['direction'] == 'bullish': contributing_factors.append('reversal_bullish') return SignalType.REVERSAL_BULLISH, contributing_factors contributing_factors.append('reversal_bearish') return SignalType.REVERSAL_BEARISH, contributing_factors # Check for range-bound market if self.is_range_bound(df, market_regime): contributing_factors.append('range_bound') return SignalType.NEUTRAL, contributing_factors # Determine signal based on total score if total_score > 0.8: contributing_factors.append('strong_buy') return SignalType.STRONG_BUY, contributing_factors elif total_score > 0.6: contributing_factors.append('buy') return SignalType.BUY, contributing_factors elif total_score < -0.8: contributing_factors.append('strong_sell') return SignalType.STRONG_SELL, contributing_factors elif total_score < -0.6: contributing_factors.append('sell') return SignalType.SELL, contributing_factors return SignalType.NO_SIGNAL, contributing_factors except Exception as e: logger.error(f"Error determining signal type: {str(e)}") return SignalType.NO_SIGNAL, [] def calculate_volume_score(self, volume_analysis: Dict[str, Any]) -> float: """Calculate volume score""" try: volume_zscore = volume_analysis.get('volume_zscore', 0.0) volume_trend = volume_analysis.get('volume_trend', 'stable') buying_pressure = volume_analysis.get('buying_pressure', 0.5) if volume_trend in ['strong_increasing', 'increasing']: trend_score = 1.0 elif volume_trend in ['strong_decreasing', 'decreasing']: trend_score = -1.0 else: trend_score = 0.0 volume_score = volume_zscore / 3 + trend_score * 0.5 + (buying_pressure - 0.5) * 2 return max(min(volume_score, 1.0), -1.0) except Exception as e: logger.error(f"Error calculating volume score: {str(e)}") return 0.0 def calculate_orderbook_score(self, orderbook_analysis: Dict[str, Any]) -> float: """Calculate orderbook score""" try: ob_pressure = orderbook_analysis.get('orderbook_pressure', {}).get('pressure', 0.0) liquidity_score = orderbook_analysis.get('liquidity_metrics', {}).get('liquidity_score', 0.0) microstructure_signals = orderbook_analysis.get('microstructure_signals', {}) # Calculate contribution from microstructure signals microstructure_score = 0.0 # Adjust the microstructure_score based on detected signals if microstructure_signals.get('bid_ask_bounce'): microstructure_score += 0.1 # Assign an appropriate value if microstructure_signals.get('buy_pressure'): microstructure_score += 0.1 if microstructure_signals.get('sell_pressure'): microstructure_score -= 0.1 if microstructure_signals.get('quote_stuffing_detected'): microstructure_score -= 0.2 if microstructure_signals.get('layers_detected'): microstructure_score -= 0.2 # Normalize microstructure_score to a reasonable range if needed microstructure_score = max(min(microstructure_score, 0.5), -0.5) # Calculate the final orderbook score with the contributions orderbook_score = ob_pressure * 0.5 + liquidity_score * 0.3 + microstructure_score * 0.2 # Ensure the score is between -1.0 and 1.0 return max(min(orderbook_score, 1.0), -1.0) except Exception as e: logger.error(f"Error calculating orderbook score: {str(e)}") return 0.0 async def calculate_success_probability( self, signal_type: SignalType, volume_analysis: Dict[str, Any], orderbook_analysis: Dict[str, Any], market_regime: MarketRegime, institutional_analysis: Dict[str, Any] ) -> float: """Calculate probability of signal success""" try: # Base probability from volume analysis volume_probability = min(1.0, volume_analysis.get('volume_zscore', 0.0) / 3) # Orderbook confidence ob_confidence = orderbook_analysis.get('liquidity_metrics', {}).get('liquidity_score', 0.0) # Institutional confidence inst_confidence = institutional_analysis.get('iceberg_probability', 0.0) # Market regime adjustment regime_multiplier = 1.0 if market_regime.regime_type == "high_volatility": regime_multiplier = 0.8 elif market_regime.regime_type == "low_liquidity": regime_multiplier = 0.7 # Combined probability base_probability = ( volume_probability * 0.4 + ob_confidence * 0.4 + inst_confidence * 0.2 ) * regime_multiplier # Adjust based on signal type if signal_type in [SignalType.STRONG_BUY, SignalType.STRONG_SELL]: base_probability *= 1.2 elif signal_type in [SignalType.REVERSAL_BULLISH, SignalType.REVERSAL_BEARISH]: base_probability *= 0.9 return min(0.95, max(0.05, base_probability)) except Exception as e: logger.error(f"Error calculating success probability: {str(e)}") return 0.5 async def calculate_risk_levels( self, signal_type: SignalType, current_price: float, orderbook_analysis: Dict[str, Any], market_impact: MarketImpact ) -> Tuple[float, float]: """Calculate stop loss and take profit levels""" try: # Base risk percentages if signal_type in [SignalType.STRONG_BUY, SignalType.STRONG_SELL]: sl_percent = 0.02 tp_percent = 0.06 elif signal_type in [SignalType.BUY, SignalType.SELL]: sl_percent = 0.015 tp_percent = 0.045 else: sl_percent = 0.025 tp_percent = 0.05 # Adjust based on market impact sl_percent *= (1 + market_impact.slippage_estimate) tp_percent *= (1 + market_impact.slippage_estimate) # Calculate levels if signal_type in [SignalType.STRONG_BUY, SignalType.BUY, SignalType.REVERSAL_BULLISH]: stop_loss = current_price * (1 - sl_percent) take_profit = current_price * (1 + tp_percent) else: stop_loss = current_price * (1 + sl_percent) take_profit = current_price * (1 - tp_percent) return stop_loss, take_profit except Exception as e: logger.error(f"Error calculating risk levels: {str(e)}") return current_price * 0.98, current_price * 1.02 def is_range_bound(self, df: pd.DataFrame, market_regime: MarketRegime) -> bool: """Determine if market is range-bound""" try: # Check volatility if market_regime.volatility > 0.02: return False recent_df = df.tail(100) price_range = (recent_df['high'].max() - recent_df['low'].min()) / recent_df['close'].mean() # Check if price is contained within a narrow range if price_range > 0.03: return False # Check for strong trends if market_regime.trend_strength > 0.02: return False # Check volume stability volume_stability = recent_df['volume'].std() / recent_df['volume'].mean() if recent_df['volume'].mean() > 0 else 0.0 if volume_stability > 0.5: return False return True except Exception as e: logger.error(f"Error checking range-bound market: {str(e)}") return False async def save_signal(self, metrics: SignalMetrics) -> None: """Save signal to database""" try: signal = Signal( timestamp=metrics.signal_time, chart_timestamp=metrics.chart_timestamp, symbol=metrics.symbol, price=metrics.current_price, price_24h_change=metrics.price_24h_change, signal=metrics.signal_type.value, stop_loss=metrics.stop_loss, take_profit=metrics.take_profit, success_rate=metrics.success_rate, result=metrics.result, analysis_details={ 'volume_analysis': metrics.volume_analysis, 'orderbook_analysis': metrics.orderbook_analysis, 'contributing_factors': metrics.contributing_factors, 'market_regime': { 'volatility': metrics.market_regime.volatility, 'trend_strength': metrics.market_regime.trend_strength, 'liquidity_score': metrics.market_regime.liquidity_score, 'regime_type': metrics.market_regime.regime_type, 'confidence': metrics.market_regime.confidence }, 'market_impact': { 'price_impact': metrics.market_impact.price_impact, 'slippage_estimate': metrics.market_impact.slippage_estimate, 'volume_impact': metrics.market_impact.volume_impact }, 'reversal_analysis': metrics.reversal_analysis, 'institutional_analysis': metrics.institutional_analysis } ) await sync_to_async(signal.save)() except Exception as e: logger.error(f"Error saving signal: {str(e)}") async def get_candles_df(self, pair: str, timeframe: str) -> pd.DataFrame: """Fetch candlestick data from database""" try: # Use cache cache_key = self.get_cache_key('candles', pair=pair, timeframe=timeframe) cached_df = await self.get_cached_data(cache_key) if cached_df is not None: return cached_df klines = await sync_to_async(list)( Kline.objects.filter( symbol=pair, timeframe=timeframe ).order_by('timestamp') .values('timestamp', 'open', 'high', 'low', 'close', 'volume') ) if not klines: return pd.DataFrame() df = pd.DataFrame(klines) df['timestamp'] = pd.to_datetime(df['timestamp']) df.set_index('timestamp', inplace=True) await self.set_cached_data(cache_key, df, self.cache_timeouts['candles']) return df except Exception as e: logger.error(f"Error fetching candles: {str(e)}") return pd.DataFrame() async def get_orderbook(self, pair: str) -> Tuple[List, List]: """Fetch orderbook data from database""" try: # Use cache cache_key = self.get_cache_key('orderbook', pair=pair) cached_orderbook = await self.get_cached_data(cache_key) if cached_orderbook is not None: return cached_orderbook orderbook = await sync_to_async(OrderBook.objects.filter(symbol=pair).first)() if not orderbook: return [], [] bids = orderbook.bids asks = orderbook.asks await self.set_cached_data(cache_key, (bids, asks), self.cache_timeouts['orderbook']) return bids, asks except Exception as e: logger.error(f"Error fetching orderbook: {str(e)}") return [], [] async def run(self): """Main method to run the bot for all symbols and timeframes.""" try: symbols = await self.get_symbols() tasks = [] for symbol in symbols: for timeframe in self.timeframes: tasks.append(self.generate_signal(symbol, timeframe)) await asyncio.gather(*tasks) except Exception as e: logger.error(f"Error running bot: {str(e)}") async def get_symbols(self) -> List[str]: """Fetch the list of trading pairs (symbols).""" trading_pairs = await sync_to_async(list)(TradingPair.objects.values_list('symbol', flat=True)) return trading_pairsif name == 'main': bot = EnhancedMarketMicrostructureBot() asyncio.run(bot.run())
Can you analyze the above code to debug it. Tell me more about institutional_presence and institutional_analysis and if they are all implemented as it should. Debug for any incomplete functions or erroneous methods. And lastly, write def calculate_institutional_score function correctly and replace the placeholder with fully implemented logic. If something needs to be modified you need to write the whole method for tthe function. - Assistant
Analyzing the code involves examining the flow, identifying any erroneous or incomplete logic, and ensuring functionalities are implemented as intended. Here’s a focused analysis of the
institutional_presenceandinstitutional_analysisfunctionalities, including a correction for thecalculate_institutional_scorefunction.Institutional Presence and Analysis
-
Purpose: The
institutional_presenceandinstitutional_analysisaim to detect significant trading behaviors, such as large orders or iceberg orders, often associated with institutional traders. -
Key Methods:
-
detect_institutional_presence: Identifies large orders and clusters them, determines iceberg possibilities, and provides smart routing suggestions based on identified patterns. -
detect_iceberg_orders: Estimates the presence of iceberg orders by assessing volume variations at the top order book levels. -
generate_smart_routing_suggestion: Provides trading strategy recommendations given the detected presence of large orders or iceberg probabilities.
-
Debugging and Enhancements
Incomplete and Erroneous Elements
-
Logic and Flow Errors:
- The
detect_institutional_presencemethod is well-defined, but the application of thresholds and interpretation logic should be reviewed to align with actual trading data dynamics. - The function
detect_iceberg_ordersassumes that variation alone indicates iceberg orders, which might not account for context or other supporting signals.
- The
-
Improving
calculate_institutional_score:- Currently, this function contains a placeholder that only considers large order counts and iceberg probability. The revised version should incorporate more nuanced factors like clustering significance and potential market impact.
Complete Implementation of
calculate_institutional_scoredef calculate_institutional_score(self, institutional_analysis: Dict[str, Any]) -> float: """Calculate the institutional score based on order characteristics and clustering.""" try: # Extract necessary information large_bids = institutional_analysis['large_orders']['bids'] large_asks = institutional_analysis['large_orders']['asks'] iceberg_prob = institutional_analysis.get('iceberg_probability', 0.0) bid_clusters = institutional_analysis['clustering']['bids'] ask_clusters = institutional_analysis['clustering']['asks'] # Define weights for scoring large_order_weight = 0.3 iceberg_weight = 0.3 clustering_weight = 0.2 market_impact_weight = 0.2 # Calculate scores large_order_score = (len(large_bids) + len(large_asks)) / 10.0 # Example normalization clustering_score = (len(bid_clusters) + len(ask_clusters)) / 10.0 # Example market impact calculation based on known impact factors market_impact_score = min(1.0, 0.1 * (len(large_bids) + len(large_asks))) # Total score computation institutional_score = ( large_order_score * large_order_weight + iceberg_prob * iceberg_weight + clustering_score * clustering_weight + market_impact_score * market_impact_weight ) # Normalize institutional_score to a range of 0 - 1 return min(institutional_score, 1.0) except Exception as e: logger.error(f"Error calculating institutional score: {str(e)}") return 0.0Suggestions for Improvements and Completeness
-
Validation: All significant operations like estimates of iceberg probability or clustering behavior should be validated against actual data patterns observed historically or via backtesting, ensuring their reliability.
-
Comprehensive Logging: Enhance logging for critical points within each function, such as calculated scores or decisions made, providing better traceability.
-
Dynamic Calibration: Consider incorporating adaptive methods that adjust weights and thresholds based on real-time feedback from model performance or backtesting results.
This analysis and enhancement plan aim to refine the functionality around institutional behavior detection, improving clarity, robustness, and adaptability in the market microstructure environment.
-
- User
Between calculate_institutional_score1, calculate_institutional_score2 and calculate_institutional_score3, wic one is better and robust def calculate_institutional_score1(self, institutional_analysis: Dict[str, Any]) -> float: """Calculate the institutional score based on order characteristics and clustering.""" try: # Extract necessary information large_bids = institutional_analysis['large_orders']['bids'] large_asks = institutional_analysis['large_orders']['asks'] iceberg_prob = institutional_analysis.get('iceberg_probability', 0.0) bid_clusters = institutional_analysis['clustering']['bids'] ask_clusters = institutional_analysis['clustering']['asks']
# Define weights for scoring large_order_weight = 0.3 iceberg_weight = 0.3 clustering_weight = 0.2 market_impact_weight = 0.2 # Calculate scores large_order_score = (len(large_bids) + len(large_asks)) / 10.0 # Example normalization clustering_score = (len(bid_clusters) + len(ask_clusters)) / 10.0 # Example market impact calculation based on known impact factors market_impact_score = min(1.0, 0.1 * (len(large_bids) + len(large_asks))) # Total score computation institutional_score = ( large_order_score * large_order_weight + iceberg_prob * iceberg_weight + clustering_score * clustering_weight + market_impact_score * market_impact_weight ) # Normalize institutional_score to a range of 0 - 1 return min(institutional_score, 1.0) except Exception as e: logger.error(f"Error calculating institutional score: {str(e)}") return 0.0 def calculate_institutional_score2(self, institutional_analysis: Dict[str, Any]) -> float: """Calculate institutional score based on various factors.""" try: large_orders_count = len(institutional_analysis['large_orders']['bids']) + len(institutional_analysis['large_orders']['asks']) iceberg_probability = institutional_analysis.get('iceberg_probability', 0.0) clustering_score = 0 # Clustering analysis: Check for concentrated large orders for side in ['bids', 'asks']: cluster_volume = sum(institutional_analysis['clustering'][side].values()) total_volume = sum(order['size'] for order in institutional_analysis['large_orders'][side]) clustering_score += (cluster_volume / total_volume) if total_volume > 0 else 0 clustering_score /= 2 # Average score from both sides # Smart routing suggestion influence smart_routing = institutional_analysis.get('smart_routing_suggestion', {}) routing_influence = 0 if smart_routing.get('execution_style') == 'careful': routing_influence = 0.2 elif smart_routing.get('execution_style') == 'aggressive': routing_influence = -0.1 # Combine factors with weights institutional_score = ( large_orders_count * 0.3 + iceberg_probability * 0.4 + clustering_score * 0.2 + routing_influence ) return max(min(institutional_score, 1.0), -1.0) # Ensure score within -1 to 1 except Exception as e: logger.error(f"Error calculating institutional score: {str(e)}") return 0.0def calculate_institutional_score3(self, institutional_analysis: Dict[str, Any]) -> float: """Calculate institutional score based on institutional_analysis data.""" try: # Sum up sizes of large orders total_large_order_volume = sum(order['size'] for order in institutional_analysis['large_orders']['bids'])
+ sum(order['size'] for order in institutional_analysis['large_orders']['asks'])# Sum up volumes in significant clusters total_cluster_volume = sum(institutional_analysis['clustering']['bids'].values()) \ + sum(institutional_analysis['clustering']['asks'].values()) # Get iceberg probability iceberg_probability = institutional_analysis.get('iceberg_probability', 0.0) # Normalize volumes (assuming the institutional threshold is significant volume) normalized_large_order_volume = min(total_large_order_volume / (10 * self.institutional_threshold), 1.0) normalized_cluster_volume = min(total_cluster_volume / (10 * self.institutional_threshold), 1.0) # Now define weights weight_large_orders = 0.4 weight_clusters = 0.4 weight_iceberg = 0.2 # Compute institutional score institutional_score = (normalized_large_order_volume * weight_large_orders + normalized_cluster_volume * weight_clusters + iceberg_probability * weight_iceberg) # Ensure the score is between 0 and 1 institutional_score = max(min(institutional_score, 1.0), 0.0) return institutional_score except Exception as e: logger.error(f"Error calculating institutional score: {str(e)}") return 0.0 - Assistant
To evaluate which version among
calculate_institutional_score1,calculate_institutional_score2, andcalculate_institutional_score3is better and more robust, let's consider several factors:Key Evaluation Metrics
- Data Utilization: The extent to which each function uses available data within
institutional_analysiseffectively. - Robustness: How well the function handles potential edge cases or anomalies in data.
- Weighting Logic: The appropriateness and rationale behind weights assigned to different factors contributing to the institutional score.
- Normalization: Ensuring scores are properly normalized and remain within reasonable bounds.
- Clear and Understandable Logic: Transparent and logical calculation processes which are easy to interpret.
Analysis
1.
calculate_institutional_score1-
Pros:
- Uses a relatively detailed list of factors: large orders, iceberg probability, clustering, and an assumed market impact.
- Introduces a straightforward and balanced weighting system.
-
Cons:
- Market impact measure (
market_impact_score) could be better defined and aligned with actual market conditions. - Normalization might not capture subtleties in very large or very small order volumes effectively.
- Market impact measure (
2.
calculate_institutional_score2-
Pros:
- Includes a detailed look into smart routing suggestions, potentially indicating response strategies to market conditions.
- Considers clustering as a calculated score based on volume concentration.
-
Cons:
- The approach to handling smart routing (
routing_influence) could lead to skewed weightings without strong behavioral indicators. - Provides a normalization range of -1 to 1, which suggests it could introduce negative factors; this could be unintended in certain calculation contexts.
- The approach to handling smart routing (
3.
calculate_institutional_score3-
Pros:
- Focused use of both large order volume and cluster volume, incorporating iceberg probability systematically.
- Effective normalization scheme using the
institutional_threshold, providing clarity in volume significance. - Ensures the score adheres to a strict 0 to 1 range, preventing unrealistic negative scores.
-
Cons:
- Simplified approach which largely depends on volumes and their direct normalization; some complexity might be warranted for nuanced market behaviors.
- Potential for slightly skewed interpretation of clustering impact due to straightforward addition.
Recommendation
Best Option:
calculate_institutional_score3- Reasoning: This version offers a precise use of the available data, focusing sharply on large orders, clustering, and iceberg probability. It manages to maintain robustness through clear normalization strategies, avoiding unintended negative scores which could misrepresent institutional presence. Moreover, the straightforward approach reduces the complexity inherent in potential misinterpretation, making it preferable for real-world applications where clear, consistent metrics are crucial.
Enhancing
calculate_institutional_score3further with dynamic adjustments based on live market feedback could continue to build on its solid foundation, ensuring adaptive responsiveness in dynamic trading environments. - Data Utilization: The extent to which each function uses available data within
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.