# source: https://raw.githubusercontent.com/PabloM03/freqtrade/a604b29b7c4f06b8d981207d8a452179ca427465/user_data/strategies/FreqAILightGBM15m.py # directory_url: https://github.com/PabloM03/freqtrade/blob/develop/user_data/strategies/ # User: PabloM03 # Repository: freqtrade # --------------------""" Github_PabloM03_freqtrade__FreqAILightGBM15m__20260827_190157 v2 — ML con LightGBM Regressor + filtros de calidad Mejoras sobre v1: - Target más largo: 96 velas (24h) en lugar de 32 (8h) → más señal, menos ruido - Entry threshold más alto: 2.5% por defecto → solo entradas de alta convicción - Filtro de tendencia triple (ema50_ok) → bloquea entradas en downtrend - HARD_TP = 0.50 → deja correr movimientos grandes (BONK, WIF, etc.) - Solo pares con historial largo: BTC, ETH, SOL (más datos = mejor entrenamiento) Walk-forward real: - train_period_days=90, backtest_period_days=30 → cero look-ahead bias Para backtesting (usar config.freqai_v2.json): freqtrade backtesting -c config.json -c config.backtest.freqai.json -c config.freqai.json -s Github_PabloM03_freqtrade__FreqAILightGBM15m__20260827_190157 --timerange 20240101-20241231 --cache none """ import logging import talib.abstract as ta import numpy as np from pandas import DataFrame from freqtrade.strategy import IStrategy, DecimalParameter logger = logging.getLogger(__name__) class Github_PabloM03_freqtrade__FreqAILightGBM15m__20260827_190157(IStrategy): INTERFACE_VERSION = 3 timeframe = "15m" startup_candle_count = 300 stoploss = -0.06 trailing_stop = False minimal_roi = {"0": 0.999} use_exit_signal = True exit_profit_only = False ignore_roi_if_entry_signal = True process_only_new_candles = True # TP duro para dejar correr grandes movimientos HARD_TP = 0.50 # Umbral de predicción — más alto = menos trades pero más calidad entry_threshold = DecimalParameter(1.5, 4.0, default=2.5, decimals=1, space="buy", optimize=True) exit_threshold = DecimalParameter(-4.0, -1.0, default=-1.5, decimals=1, space="sell", optimize=True) # Filtro de tendencia (ema50_ok) buy_ema50_close_pct = DecimalParameter(0.920, 0.998, default=0.970, decimals=3, space="buy", optimize=True) buy_ema50_slope = DecimalParameter(0.960, 0.998, default=0.985, decimals=3, space="buy", optimize=True) def feature_engineering_expand_all( self, dataframe: DataFrame, period: int, metadata: dict, **kwargs ) -> DataFrame: """ Features por cada período en indicator_periods_candles [10, 20, 40]. Naming literal "period" — FreqAI renombra internamente. """ dataframe["%-rsi-period"] = ta.RSI(dataframe, timeperiod=period) dataframe["%-adx-period"] = ta.ADX(dataframe, timeperiod=period) dataframe["%-mfi-period"] = ta.MFI(dataframe, timeperiod=period) dataframe["%-roc-period"] = ta.ROC(dataframe, timeperiod=period) dataframe["%-macdhist-period"] = ta.MACD(dataframe)["macdhist"] dataframe["%-willr-period"] = ta.WILLR(dataframe, timeperiod=period) dataframe["%-cci-period"] = ta.CCI(dataframe, timeperiod=period) bb = ta.BBANDS(dataframe, timeperiod=period) upper = bb["upperband"] lower = bb["lowerband"] mid = bb["middleband"] dataframe["%-bb_pct-period"] = (dataframe["close"] - lower) / (upper - lower + 1e-8) dataframe["%-bb_width-period"] = (upper - lower) / (mid + 1e-8) return dataframe def feature_engineering_expand_basic( self, dataframe: DataFrame, metadata: dict, **kwargs ) -> DataFrame: """Features por cada timeframe incluido.""" dataframe["%-pct_1"] = dataframe["close"].pct_change(1) dataframe["%-pct_4"] = dataframe["close"].pct_change(4) dataframe["%-pct_16"] = dataframe["close"].pct_change(16) dataframe["%-pct_32"] = dataframe["close"].pct_change(32) dataframe["%-vol_ratio"] = dataframe["volume"] / (dataframe["volume"].rolling(20).mean() + 1e-8) dataframe["%-vol_ratio_4"] = dataframe["volume"] / (dataframe["volume"].rolling(4).mean() + 1e-8) dataframe["%-raw_price"] = dataframe["close"] dataframe["%-raw_volume"] = dataframe["volume"] return dataframe def feature_engineering_standard( self, dataframe: DataFrame, metadata: dict, **kwargs ) -> DataFrame: """Features globales + TARGET.""" ema200 = ta.EMA(dataframe, timeperiod=200) ema80 = ta.EMA(dataframe, timeperiod=80) ema50 = ta.EMA(dataframe, timeperiod=50) dataframe["%-close_ema200"] = dataframe["close"] / (ema200 + 1e-8) dataframe["%-close_ema80"] = dataframe["close"] / (ema80 + 1e-8) dataframe["%-close_ema50"] = dataframe["close"] / (ema50 + 1e-8) dataframe["%-ema50_ema200"] = ema50 / (ema200 + 1e-8) dataframe["%-ema200_slope"] = ema200 / (ema200.shift(96) + 1e-8) dataframe["%-ema50_slope"] = ema50 / (ema50.shift(32) + 1e-8) atr = ta.ATR(dataframe, timeperiod=14) dataframe["%-atr_ratio"] = atr / (dataframe["close"] + 1e-8) stoch = ta.STOCHRSI(dataframe, timeperiod=14, fastk_period=3, fastd_period=3) dataframe["%-stoch_k"] = stoch["fastk"] dataframe["%-stoch_d"] = stoch["fastd"] # TARGET: % de cambio en las próximas 96 velas (24h a 15m) — más señal que 8h dataframe["&-s_target"] = ( dataframe["close"].shift(-96) / dataframe["close"] - 1 ) * 100 # Para filtro ema50_ok (no es un feature ML — se usa solo en populate_entry_trend) dataframe["ema50_ht"] = ema80 # EMA80@15m = 20h — "ema_fast" equivalente dataframe["ema200_ht"] = ema200 # EMA200@15m = 50h — "ema_slow" return dataframe def populate_indicators(self, dataframe: DataFrame, metadata: dict) -> DataFrame: dataframe = self.freqai.start(dataframe, metadata, self) return dataframe def populate_entry_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame: # Filtro de tendencia: EMA200 no bajando en 48h + precio cerca de EMA200 ema50_ok = ( (dataframe["ema200_ht"] >= dataframe["ema200_ht"].shift(192) * self.buy_ema50_slope.value) & (dataframe["close"] >= dataframe["ema200_ht"] * self.buy_ema50_close_pct.value) ) conditions = [ dataframe["&-s_target"] > self.entry_threshold.value, dataframe["do_predict"] == 1, ema50_ok, dataframe["volume"] > 0, ] dataframe.loc[np.all(conditions, axis=0), "enter_long"] = 1 return dataframe def populate_exit_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame: conditions = [ dataframe["&-s_target"] < self.exit_threshold.value, dataframe["do_predict"] == 1, dataframe["volume"] > 0, ] dataframe.loc[np.all(conditions, axis=0), "exit_long"] = 1 return dataframe def custom_exit(self, pair, trade, current_time, current_rate, current_profit, **kwargs): # noqa: ARG002 """Hard TP para dejar correr grandes movimientos.""" if current_profit is not None and current_profit >= self.HARD_TP: return "hard_tp" return None