# source: https://raw.githubusercontent.com/raphant/freqgym/7c3d8146c30a24d8b8e866793a2456b531986de4/user_data/strategies/FreqGym_normalized.py # pragma pylint: disable=missing-docstring, invalid-name, pointless-string-statement import logging import time from pathlib import Path import freqtrade.vendor.qtpylib.indicators as qtpylib import numpy as np # noqa import pandas as pd # noqa import talib as ta import talib.abstract as ta from freqtrade.strategy.interface import IStrategy from pandas import DataFrame from stable_baselines3 import A2C from stable_baselines3.ppo.ppo import PPO logger = logging.getLogger(__name__) COLUMNS_FILTER = [ "date", "open", "close", "high", "low", "volume", "buy", "sell", "buy_tag", "exit_tag", ] class github_raphant_freqgym__FreqGym_normalized__20230208_211859(IStrategy): # # If you've used SimpleROIEnv then use this minimal_roi # minimal_roi = { # "720": -10, # "600": 0.00001, # "60": 0.01, # "30": 0.02, # "0": 0.03 # } minimal_roi = {"0": 100} stoploss = -0.99 # Trailing stop: trailing_stop = False trailing_stop_positive = 0.001 trailing_stop_positive_offset = 0.017 trailing_only_offset_is_reached = True ticker_interval = "5m" use_sell_signal = True # Run "populate_indicators()" only for new candle. process_only_new_candles = False startup_candle_count: int = 200 model = None window_size = None timeperiods = [7, 14, 21] def __init__(self, config: dict) -> None: super().__init__(config) try: # get the file that starts with "best_model_" in the models/ directory # list files in the directory # files = Path('models/').glob('final_model_*') # get the first file # model_file = next(files) model_file = Path( "models/best_model_github_raphant_freqgym__FreqGym_normalized__20230208_211859_FreqtradeEnv_A2C_20220318_115558.zip" ) assert model_file.exists(), f'Model file "{model_file}" does not exist.' self.model = A2C.load( str(model_file) ) # Note: Make sure you use the same policy as the one used to train self.window_size = self.model.observation_space.shape[0] except Exception as e: logger.exception(f"Could not load model: {e}") else: logger.info(f"Loaded model: {model_file}") def populate_indicators(self, dataframe: DataFrame, metadata: dict) -> DataFrame: """ Adds several different TA indicators to the given DataFrame Performance Note: For the best performance be frugal on the number of indicators you are using. Let uncomment only the indicator you are using in your strategies or your hyperopt configuration, otherwise you will waste your memory and CPU usage. :param dataframe: Raw data from the exchange and parsed by parse_ticker_dataframe() :param metadata: Additional information, like the currently traded pair :return: a Dataframe with all mandatory indicators for the strategies """ logger.info(f'Calculating TA indicators for {metadata["pair"]}') # Plus Directional Indicator / Movement dataframe["plus_di"] = normalize(ta.PLUS_DI(dataframe), 0, 100) # # Minus Directional Indicator / Movement dataframe["minus_di"] = normalize(ta.MINUS_DI(dataframe), 0, 100) # Ultimate Oscillator dataframe["uo"] = normalize(ta.ULTOSC(dataframe), 0, 100) # Hilbert Transform Indicator - SineWave hilbert = ta.HT_SINE(dataframe) dataframe["htsine"] = normalize(hilbert["sine"], -1, 1) dataframe["htleadsine"] = normalize(hilbert["leadsine"], -1, 1) # BOP Balance Of Power dataframe["bop"] = normalize(ta.BOP(dataframe), -1, 1) # STOCH - Stochastic stoch = ta.STOCH(dataframe) dataframe["slowk"] = normalize(stoch["slowk"], 0, 100) dataframe["slowd"] = normalize(stoch["slowd"], 0, 100) # STOCHF - Stochastic Fast stochf = ta.STOCHF(dataframe) dataframe["fastk"] = normalize(stochf["fastk"], 0, 100) dataframe["fastk"] = normalize(stochf["fastk"], 0, 100) # Bollinger Bands bollinger2 = qtpylib.bollinger_bands( qtpylib.typical_price(dataframe), window=20, stds=2 ) bollinger3 = qtpylib.bollinger_bands( qtpylib.typical_price(dataframe), window=20, stds=3 ) dataframe["bb2_lower_gt_close"] = ( bollinger2["lower"].gt(dataframe["close"]).astype("int") ) dataframe["bb3_lower_gt_close"] = ( bollinger3["lower"].gt(dataframe["close"]).astype("int") ) for period in self.timeperiods: # ADX Average Directional Movement Index dataframe[f"adx_{period}"] = normalize( ta.ADX(dataframe, timeperiod=period), 0, 100 ) # Aroon, Aroon Oscillator aroon = ta.AROON(dataframe, timeperiod=period) dataframe[f"aroonup_{period}"] = normalize(aroon["aroonup"], 0, 100) dataframe[f"aroondown_{period}"] = normalize(aroon["aroondown"], 0, 100) dataframe[f"aroonosc_{period}"] = normalize( ta.AROONOSC(dataframe, timeperiod=period), -100, 100 ) # CMO Chande Momentum Oscillator dataframe[f"cmo_{period}"] = normalize( ta.CMO(dataframe, timeperiod=period), -100, 100 ) # DX Directional Movement Index dataframe[f"dx_{period}"] = normalize( ta.DX(dataframe, timeperiod=period), 0, 100 ) # MFI Money Flow Index dataframe[f"mfi_{period}"] = normalize( ta.MFI(dataframe, timeperiod=period), 0, 100 ) # MINUS_DI Minus Directional Indicator dataframe[f"minus_di_{period}"] = normalize( ta.MINUS_DI(dataframe, timeperiod=period), 0, 100 ) # PLUS_DI Plus Directional Indicator dataframe[f"plus_di_{period}"] = normalize( ta.PLUS_DI(dataframe, timeperiod=period), 0, 100 ) # Williams %R dataframe[f"willr_{period}"] = normalize( ta.WILLR(dataframe, timeperiod=period), -100, 0 ) # RSI dataframe[f"rsi_{period}"] = normalize( ta.RSI(dataframe, timeperiod=period), 0, 100 ) # Inverse Fisher transform on RSI: values [-1.0, 1.0] (https://goo.gl/2JGGoy) rsi = 0.1 * (dataframe[f"rsi_{period}"] - 50) dataframe[f"fisher_rsi_{period}"] = (np.exp(2 * rsi) - 1) / ( np.exp(2 * rsi) + 1 ) dataframe[f"fisher_rsi_{period}"] = normalize( dataframe[f"fisher_rsi_{period}"], -1, 1 ) # STOCHRSI - Stochastic Relative Strength Index stoch_rsi = ta.STOCHRSI(dataframe, timeperiod=period) dataframe[f"stochrsi_k_{period}"] = normalize(stoch_rsi["fastk"], 0, 100) dataframe[f"stochrsi_d_{period}"] = normalize(stoch_rsi["fastd"], 0, 100) # # CORREL - Pearson's Correlation Coefficient (r) # dataframe[f'correl_{period}'] = normalize(ta.CORREL(dataframe, timeperiod=period), -1, 1) # this is buggy # LINEARREG_ANGLE - Linear Regression Angle dataframe[f"linangle_{period}"] = normalize( ta.LINEARREG_ANGLE(dataframe, timeperiod=period), -90, 90 ) indicators = dataframe[ dataframe.columns[~dataframe.columns.isin(COLUMNS_FILTER)] ] assert all(indicators.max() < 1.00001) and all( indicators.min() > -0.00001 ), "Error, values are not normalized!" logger.info(f'{metadata["pair"]} - indicators populated!') return dataframe def populate_buy_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame: """ Based on TA indicators, populates the buy signal for the given dataframe :param dataframe: DataFrame populated with indicators :param metadata: Additional information, like the currently traded pair :return: DataFrame with buy column """ # dataframe['buy'] = self.rl_model_predict(dataframe) logger.info(f'Populating buy signal for {metadata["pair"]}') action = self.rl_model_predict(dataframe) dataframe["buy"] = (action == 1).astype("int") logger.info(f'{metadata["pair"]} - buy signal populated!') return dataframe def populate_sell_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame: """ Based on TA indicators, populates the sell signal for the given dataframe :param dataframe: DataFrame populated with indicators :param metadata: Additional information, like the currently traded pair :return: DataFrame with buy column """ logger.info(f'Populating sell signal for {metadata["pair"]}') action = self.rl_model_predict(dataframe) dataframe["sell"] = (action == 2).astype("int") logger.info(f'{metadata["pair"]} - sell signal populated!') return dataframe def rl_model_predict(self, dataframe: pd.DataFrame) -> pd.DataFrame: output = pd.DataFrame(np.zeros((len(dataframe), 1))) indicators = ( dataframe[dataframe.columns[~dataframe.columns.isin(COLUMNS_FILTER)]] .fillna(0) .to_numpy() ) indicators2 = dataframe[ dataframe.columns[~dataframe.columns.isin(COLUMNS_FILTER)] ].fillna(0) # convert above code to rolling # self.model.predict expects a numpy array of shape (batch_size, seq_len) # t1 = time.time() # second_output = indicators2.rolling( # window=self.window_size, min_periods=self.window_size # ).apply(lambda x: self.predict(x)) # logger.info(f'Elapsed time rolling: {time.time() - t1}') # TODO: This is slow and ugly, must use .rolling t1 = time.time() for window in range(self.window_size, len(dataframe)): start = window - self.window_size end = window observation = indicators[start:end] res, _ = predict.predict(observation, deterministic=True) output.loc[end] = res logger.info(f"Elapsed time default: {time.time() - t1}") return output def predict(self, x: pd.Series): logger.info(f"Predicting action for {x}, type: {type(x)}") return predict.predict(x, deterministic=True) def normalize(data, min_value, max_value): """ Given a list of numbers, normalize each number by subtracting the minimum and dividing by the maximum :param data: The data to be normalized :param min_value: the minimum value in the data :param max_value: The maximum value in the column :return: The normalized data """ return (data - min_value) / (max_value - min_value)