from typing import Dict, List, NamedTuple, Optional import arrow from pandas import DataFrame from freqtrade.enums import SellType from freqtrade.exchange import timeframe_to_minutes tests_start_time = arrow.get(2018, 10, 3) tests_timeframe = '1h' class BTrade(NamedTuple): """ Minimalistic Trade result used for functional backtesting """ sell_reason: SellType open_tick: int close_tick: int buy_tag: Optional[str] = None class BTContainer(NamedTuple): """ Minimal BacktestContainer defining Backtest inputs and results. """ data: List[List[float]] stop_loss: float roi: Dict[str, float] trades: List[BTrade] profit_perc: float trailing_stop: bool = False trailing_only_offset_is_reached: bool = False trailing_stop_positive: Optional[float] = None trailing_stop_positive_offset: float = 0.0 use_sell_signal: bool = False use_custom_stoploss: bool = False def _get_frame_time_from_offset(offset): minutes = offset * timeframe_to_minutes(tests_timeframe) return tests_start_time.shift(minutes=minutes).datetime def _build_backtest_dataframe(data): columns = ['date', 'open', 'high', 'low', 'close', 'volume', 'buy', 'sell'] columns = columns + ['buy_tag'] if len(data[0]) == 9 else columns frame = DataFrame.from_records(data, columns=columns) frame['date'] = frame['date'].apply(_get_frame_time_from_offset) # Ensure floats are in place for column in ['open', 'high', 'low', 'close', 'volume']: frame[column] = frame[column].astype('float64') if 'buy_tag' not in columns: frame['buy_tag'] = None # Ensure all candles make kindof sense assert all(frame['low'] <= frame['close']) assert all(frame['low'] <= frame['open']) assert all(frame['high'] >= frame['close']) assert all(frame['high'] >= frame['open']) return frame