lib.py

"""
Collection of common building blocks, helper auxiliary functions and
composable strategy classes for reuse.

Intended for simple missing-link procedures, not reinventing
of better-suited, state-of-the-art, fast libraries,
such as TA-Lib, Tulipy, PyAlgoTrade, NumPy, SciPy ...

Please raise ideas for additions to this collection on the [issue tracker].

[issue tracker]: https://github.com/kernc/backtesting.py
"""

from __future__ import annotations

import multiprocessing as mp
from collections import OrderedDict
from inspect import currentframe
from itertools import chain, compress, count
from numbers import Number
from typing import Callable, Generator, Optional, Sequence, Union

import numpy as np
import pandas as pd

from ._plotting import plot_heatmaps as _plot_heatmaps
from ._stats import compute_stats as _compute_stats
from ._util import SharedMemoryManager, _Array, _as_str, _batch, _tqdm
from .backtesting import Backtest, Strategy

__pdoc__ = {}


OHLCV_AGG = OrderedDict((
    ('Open', 'first'),
    ('High', 'max'),
    ('Low', 'min'),
    ('Close', 'last'),
    ('Volume', 'sum'),
))
"""Dictionary of rules for aggregating resampled OHLCV data frames,
e.g.

    df.resample('4H', label='right').agg(OHLCV_AGG).dropna()
"""

TRADES_AGG = OrderedDict((
    ('Size', 'sum'),
    ('EntryBar', 'first'),
    ('ExitBar', 'last'),
    ('EntryPrice', 'mean'),
    ('ExitPrice', 'mean'),
    ('PnL', 'sum'),
    ('ReturnPct', 'mean'),
    ('EntryTime', 'first'),
    ('ExitTime', 'last'),
    ('Duration', 'sum'),
))
"""Dictionary of rules for aggregating resampled trades data,
e.g.

    stats['_trades'].resample('1D', on='ExitTime',
                              label='right').agg(TRADES_AGG)
"""

_EQUITY_AGG = {
    'Equity': 'last',
    'DrawdownPct': 'max',
    'DrawdownDuration': 'max',
}


def barssince(condition: Sequence[bool], default=np.inf) -> int:
    """
    Return the number of bars since `condition` sequence was last `True`,
    or if never, return `default`.

        >>> barssince(self.data.Close > self.data.Open)
        3
    """
    return next(compress(range(len(condition)), reversed(condition)), default)


def cross(series1: Sequence, series2: Sequence) -> bool:
    """
    Return `True` if `series1` and `series2` just crossed
    (above or below) each other.

        >>> cross(self.data.Close, self.sma)
        True

    """
    return crossover(series1, series2) or crossover(series2, series1)


def crossover(series1: Sequence, series2: Sequence) -> bool:
    """
    Return `True` if `series1` just crossed over (above)
    `series2`.

        >>> crossover(self.data.Close, self.sma)
        True
    """
    series1 = (
        series1.values if isinstance(series1, pd.Series) else
        (series1, series1) if isinstance(series1, Number) else
        series1)
    series2 = (
        series2.values if isinstance(series2, pd.Series) else
        (series2, series2) if isinstance(series2, Number) else
        series2)
    try:
        return series1[-2] < series2[-2] and series1[-1] > series2[-1]  # type: ignore
    except IndexError:
        return False


def plot_heatmaps(heatmap: pd.Series,
                  agg: Union[str, Callable] = 'max',
                  *,
                  ncols: int = 3,
                  plot_width: int = 1200,
                  filename: str = '',
                  open_browser: bool = True):
    """
    Plots a grid of heatmaps, one for every pair of parameters in `heatmap`.
    See example in [the tutorial].

    [the tutorial]: https://kernc.github.io/backtesting.py/doc/examples/Parameter%20Heatmap%20&%20Optimization.html#plot-heatmap  # noqa: E501

    `heatmap` is a Series as returned by
    `backtesting.backtesting.Backtest.optimize` when its parameter
    `return_heatmap=True`.

    When projecting the n-dimensional (n > 2) heatmap onto 2D, the values are
    aggregated by 'max' function by default. This can be tweaked
    with `agg` parameter, which accepts any argument pandas knows
    how to aggregate by.

    .. todo::
        Lay heatmaps out lower-triangular instead of in a simple grid.
        Like [`sambo.plot.plot_objective()`][plot_objective] does.

    [plot_objective]: \
        https://sambo-optimization.github.io/doc/sambo/plot.html#sambo.plot.plot_objective
    """
    return _plot_heatmaps(heatmap, agg, ncols, filename, plot_width, open_browser)


def quantile(series: Sequence, quantile: Union[None, float] = None):
    """
    If `quantile` is `None`, return the quantile _rank_ of the last
    value of `series` wrt former series values.

    If `quantile` is a value between 0 and 1, return the _value_ of
    `series` at this quantile. If used to working with percentiles, just
    divide your percentile amount with 100 to obtain quantiles.

        >>> quantile(self.data.Close[-20:], .1)
        162.130
        >>> quantile(self.data.Close)
        0.13
    """
    if quantile is None:
        try:
            last, series = series[-1], series[:-1]
            return np.mean(series < last)
        except IndexError:
            return np.nan
    assert 0 <= quantile <= 1, "quantile must be within [0, 1]"
    return np.nanpercentile(series, quantile * 100)


def compute_stats(
        *,
        stats: pd.Series,
        data: pd.DataFrame,
        trades: pd.DataFrame = None,
        risk_free_rate: float = 0.) -> pd.Series:
    """
    (Re-)compute strategy performance metrics.

    `stats` is the statistics series as returned by `backtesting.backtesting.Backtest.run()`.
    `data` is OHLC data as passed to the `backtesting.backtesting.Backtest`
    the `stats` were obtained in.
    `trades` can be a dataframe subset of `stats._trades` (e.g. only long trades).
    You can also tune `risk_free_rate`, used in calculation of Sharpe and Sortino ratios.

        >>> stats = Backtest(GOOG, MyStrategy).run()
        >>> only_long_trades = stats._trades[stats._trades.Size > 0]
        >>> long_stats = compute_stats(stats=stats, trades=only_long_trades,
        ...                            data=GOOG, risk_free_rate=.02)
    """
    equity = stats._equity_curve.Equity
    if trades is None:
        trades = stats._trades
    else:
        # XXX: Is this buggy?
        equity = equity.copy()
        equity[:] = stats._equity_curve.Equity.iloc[0]
        for t in trades.itertuples(index=False):
            equity.iloc[t.EntryBar:] += t.PnL
    return _compute_stats(trades=trades, equity=equity.values, ohlc_data=data,
                          risk_free_rate=risk_free_rate, strategy_instance=stats._strategy)


def resample_apply(rule: str,
                   func: Optional[Callable[..., Sequence]],
                   series: Union[pd.Series, pd.DataFrame, _Array],
                   *args,
                   agg: Optional[Union[str, dict]] = None,
                   **kwargs):
    """
    Apply `func` (such as an indicator) to `series`, resampled to
    a time frame specified by `rule`. When called from inside
    `backtesting.backtesting.Strategy.init`,
    the result (returned) series will be automatically wrapped in
    `backtesting.backtesting.Strategy.I`
    wrapper method.

    `rule` is a valid [Pandas offset string] indicating
    a time frame to resample `series` to.

    [Pandas offset string]: \
http://pandas.pydata.org/pandas-docs/stable/timeseries.html#offset-aliases

    `func` is the indicator function to apply on the resampled series.

    `series` is a data series (or array), such as any of the
    `backtesting.backtesting.Strategy.data` series. Due to pandas
    resampling limitations, this only works when input series
    has a datetime index.

    `agg` is the aggregation function to use on resampled groups of data.
    Valid values are anything accepted by `pandas/resample/.agg()`.
    Default value for dataframe input is `OHLCV_AGG` dictionary.
    Default value for series input is the appropriate entry from `OHLCV_AGG`
    if series has a matching name, or otherwise the value `"last"`,
    which is suitable for closing prices,
    but you might prefer another (e.g. `"max"` for peaks, or similar).

    Finally, any `*args` and `**kwargs` that are not already eaten by
    implicit `backtesting.backtesting.Strategy.I` call
    are passed to `func`.

    For example, if we have a typical moving average function
    `SMA(values, lookback_period)`, _hourly_ data source, and need to
    apply the moving average MA(10) on a _daily_ time frame,
    but don't want to plot the resulting indicator, we can do:

        class System(Strategy):
            def init(self):
                self.sma = resample_apply(
                    'D', SMA, self.data.Close, 10, plot=False)

    The above short snippet is roughly equivalent to:

        class System(Strategy):
            def init(self):
                # Strategy exposes `self.data` as raw NumPy arrays.
                # Let's convert closing prices back to pandas Series.
                close = self.data.Close.s

                # Resample to daily resolution. Aggregate groups
                # using their last value (i.e. closing price at the end
                # of the day). Notice `label='right'`. If it were set to
                # 'left' (default), the strategy would exhibit
                # look-ahead bias.
                daily = close.resample('D', label='right').agg('last')

                # We apply SMA(10) to daily close prices,
                # then reindex it back to original hourly index,
                # forward-filling the missing values in each day.
                # We make a separate function that returns the final
                # indicator array.
                def SMA(series, n):
                    from backtesting.test import SMA
                    return SMA(series, n).reindex(close.index).ffill()

                # The result equivalent to the short example above:
                self.sma = self.I(SMA, daily, 10, plot=False)

    """
    if func is None:
        def func(x, *_, **__):
            return x
    assert callable(func), 'resample_apply(func=) must be callable'

    if not isinstance(series, (pd.Series, pd.DataFrame)):
        assert isinstance(series, _Array), \
            'resample_apply(series=) must be `pd.Series`, `pd.DataFrame`, ' \
            'or a `Strategy.data.*` array'
        series = series.s

    if agg is None:
        agg = OHLCV_AGG.get(getattr(series, 'name', ''), 'last')
        if isinstance(series, pd.DataFrame):
            agg = {column: OHLCV_AGG.get(column, 'last')
                   for column in series.columns}

    resampled = series.resample(rule, label='right').agg(agg).dropna()
    resampled.name = _as_str(series) + '[' + rule + ']'

    # Check first few stack frames if we are being called from
    # inside Strategy.init, and if so, extract Strategy.I wrapper.
    frame, level = currentframe(), 0
    while frame and level <= 3:
        frame = frame.f_back
        level += 1
        if isinstance(frame.f_locals.get('self'), Strategy):  # type: ignore
            strategy_I = frame.f_locals['self'].I             # type: ignore
            break
    else:
        def strategy_I(func, *args, **kwargs):  # noqa: F811
            return func(*args, **kwargs)

    def wrap_func(resampled, *args, **kwargs):
        result = func(resampled, *args, **kwargs)
        if not isinstance(result, pd.DataFrame) and not isinstance(result, pd.Series):
            result = np.asarray(result)
            if result.ndim == 1:
                result = pd.Series(result, name=resampled.name)
            elif result.ndim == 2:
                result = pd.DataFrame(result.T)
        # Resample back to data index
        if not isinstance(result.index, pd.DatetimeIndex):
            result.index = resampled.index
        result = result.reindex(index=series.index.union(resampled.index),
                                method='ffill').reindex(series.index)
        return result

    wrap_func.__name__ = func.__name__

    array = strategy_I(wrap_func, resampled, *args, **kwargs)
    return array


def random_ohlc_data(example_data: pd.DataFrame, *,
                     frac=1., random_state: Optional[int] = None) -> Generator[pd.DataFrame, None, None]:
    """
    OHLC data generator. The generated OHLC data has basic
    [descriptive statistics](https://en.wikipedia.org/wiki/Descriptive_statistics)
    similar to the provided `example_data`.

    `frac` is a fraction of data to sample (with replacement). Values greater
    than 1 result in oversampling.

    Such random data can be effectively used for stress testing trading
    strategy robustness, Monte Carlo simulations, significance testing, etc.

    >>> from backtesting.test import EURUSD
    >>> ohlc_generator = random_ohlc_data(EURUSD)
    >>> next(ohlc_generator)  # returns new random data
    ...
    >>> next(ohlc_generator)  # returns new random data
    ...
    """
    def shuffle(x):
        return x.sample(frac=frac, replace=frac > 1, random_state=random_state)

    if len(example_data.columns.intersection({'Open', 'High', 'Low', 'Close'})) != 4:
        raise ValueError("`data` must be a pandas.DataFrame with columns "
                         "'Open', 'High', 'Low', 'Close'")
    while True:
        df = shuffle(example_data)
        df.index = example_data.index
        padding = df.Close - df.Open.shift(-1)
        gaps = shuffle(example_data.Open.shift(-1) - example_data.Close)
        deltas = (padding + gaps).shift(1).fillna(0).cumsum()
        for key in ('Open', 'High', 'Low', 'Close'):
            df[key] += deltas
        yield df


class SignalStrategy(Strategy):
    """
    A simple helper strategy that operates on position entry/exit signals.
    This makes the backtest of the strategy simulate a [vectorized backtest].
    See [tutorials] for usage examples.

    [vectorized backtest]: https://www.google.com/search?q=vectorized+backtest
    [tutorials]: index.html#tutorials

    To use this helper strategy, subclass it, override its
    `backtesting.backtesting.Strategy.init` method,
    and set the signal vector by calling
    `backtesting.lib.SignalStrategy.set_signal` method from within it.

        class ExampleStrategy(SignalStrategy):
            def init(self):
                super().init()
                self.set_signal(sma1 > sma2, sma1 < sma2)

    Remember to call `super().init()` and `super().next()` in your
    overridden methods.
    """
    __entry_signal = (0,)
    __exit_signal = (False,)

    def set_signal(self, entry_size: Sequence[float],
                   exit_portion: Optional[Sequence[float]] = None,
                   *,
                   plot: bool = True):
        """
        Set entry/exit signal vectors (arrays).

        A long entry signal is considered present wherever `entry_size`
        is greater than zero, and a short signal wherever `entry_size`
        is less than zero, following `backtesting.backtesting.Order.size` semantics.

        If `exit_portion` is provided, a nonzero value closes portion the position
        (see `backtesting.backtesting.Trade.close()`) in the respective direction
        (positive values close long trades, negative short).

        If `plot` is `True`, the signal entry/exit indicators are plotted when
        `backtesting.backtesting.Backtest.plot` is called.
        """
        self.__entry_signal = self.I(  # type: ignore
            lambda: pd.Series(entry_size, dtype=float).replace(0, np.nan),
            name='entry size', plot=plot, overlay=False, scatter=True, color='black')

        if exit_portion is not None:
            self.__exit_signal = self.I(  # type: ignore
                lambda: pd.Series(exit_portion, dtype=float).replace(0, np.nan),
                name='exit portion', plot=plot, overlay=False, scatter=True, color='black')

    def next(self):
        super().next()

        exit_portion = self.__exit_signal[-1]
        if exit_portion > 0:
            for trade in self.trades:
                if trade.is_long:
                    trade.close(exit_portion)
        elif exit_portion < 0:
            for trade in self.trades:
                if trade.is_short:
                    trade.close(-exit_portion)

        entry_size = self.__entry_signal[-1]
        if entry_size > 0:
            self.buy(size=entry_size)
        elif entry_size < 0:
            self.sell(size=-entry_size)


class TrailingStrategy(Strategy):
    """
    A strategy with automatic trailing stop-loss, trailing the current
    price at distance of some multiple of average true range (ATR). Call
    `TrailingStrategy.set_trailing_sl()` to set said multiple
    (`6` by default). See [tutorials] for usage examples.

    [tutorials]: index.html#tutorials

    Remember to call `super().init()` and `super().next()` in your
    overridden methods.
    """
    __n_atr = 6.
    __atr = None

    def init(self):
        super().init()
        self.set_atr_periods()

    def set_atr_periods(self, periods: int = 100):
        """
        Set the lookback period for computing ATR. The default value
        of 100 ensures a _stable_ ATR.
        """
        hi, lo, c_prev = self.data.High, self.data.Low, pd.Series(self.data.Close).shift(1)
        tr = np.max([hi - lo, (c_prev - hi).abs(), (c_prev - lo).abs()], axis=0)
        atr = pd.Series(tr).ffill().bfill().rolling(periods).mean().values
        self.__atr = atr

    def set_trailing_sl(self, n_atr: float = 6):
        """
        Set the future trailing stop-loss as some multiple (`n_atr`)
        average true bar ranges away from the current price.
        """
        self.__n_atr = n_atr

    def set_trailing_pct(self, pct: float = .05):
        """
        Set the future trailing stop-loss as some percent (`0 < pct < 1`)
        below the current price (default 5% below).

        .. note:: Stop-loss set by `pct` is inexact
            Stop-loss set by `set_trailing_pct` is converted to units of ATR
            with `mean(Close * pct / atr)` and set with `set_trailing_sl`.
        """
        assert 0 < pct < 1, 'Need pct= as rate, i.e. 5% == 0.05'
        pct_in_atr = np.mean(self.data.Close * pct / self.__atr)  # type: ignore
        self.set_trailing_sl(pct_in_atr)

    def next(self):
        super().next()
        # Can't use index=-1 because self.__atr is not an Indicator type
        index = len(self.data)-1
        for trade in self.trades:
            if trade.is_long:
                trade.sl = max(trade.sl or -np.inf,
                               self.data.Close[index] - self.__atr[index] * self.__n_atr)
            else:
                trade.sl = min(trade.sl or np.inf,
                               self.data.Close[index] + self.__atr[index] * self.__n_atr)


class FractionalBacktest(Backtest):
    """
    A `backtesting.backtesting.Backtest` that supports fractional share trading
    by simple composition. It applies roughly the transformation:

        df = (df / satoshi).assign(Volume=df.Volume * satoshi)

    as unchallenged in [this FAQ entry on GitHub](https://github.com/kernc/backtesting.py/issues/134),
    then passes `data`, `args*`, and `**kwargs` to its super.

    Parameter `satoshi` tells the amount of scaling to do. E.g. for
    μBTC trading, pass `satoshi=1e6`.
    """
    def __init__(self,
                 data,
                 *args,
                 satoshi=int(100e6),
                 **kwargs):
        data = data.copy()
        data[['Open', 'High', 'Low', 'Close']] /= satoshi
        data['Volume'] *= satoshi
        super().__init__(data, *args, **kwargs)


# Prevent pdoc3 documenting __init__ signature of Strategy subclasses
for cls in list(globals().values()):
    if isinstance(cls, type) and issubclass(cls, Strategy):
        __pdoc__[f'{cls.__name__}.__init__'] = False


class MultiBacktest:
    """
    Multi-dataset `backtesting.backtesting.Backtest` wrapper.

    Run supplied `backtesting.backtesting.Strategy` on several instruments,
    in parallel.  Used for comparing strategy runs across many instruments
    or classes of instruments. Example:

        from backtesting.test import EURUSD, BTCUSD, SmaCross
        btm = MultiBacktest([EURUSD, BTCUSD], SmaCross)
        stats_per_ticker: pd.DataFrame = btm.run(fast=10, slow=20)
        heatmap_per_ticker: pd.DataFrame = btm.optimize(...)
    """
    def __init__(self, df_list, strategy_cls, **kwargs):
        self._dfs = df_list
        self._strategy = strategy_cls
        self._bt_kwargs = kwargs

    def run(self, **kwargs):
        """
        Wraps `backtesting.backtesting.Backtest.run`. Returns `pd.DataFrame` with
        currency indexes in columns.
        """
        with mp.Pool() as pool, \
                SharedMemoryManager() as smm:
            shm = [smm.df2shm(df) for df in self._dfs]
            results = _tqdm(
                pool.imap(self._mp_task_run,
                          ((df_batch, self._strategy, self._bt_kwargs, kwargs)
                           for df_batch in _batch(shm))),
                total=len(shm),
                desc=self.__class__.__name__,
            )
            df = pd.DataFrame(list(chain(*results))).transpose()
        return df

    @staticmethod
    def _mp_task_run(args):
        data_shm, strategy, bt_kwargs, run_kwargs = args
        dfs, shms = zip(*(SharedMemoryManager.shm2df(i) for i in data_shm))
        try:
            return [stats.filter(regex='^[^_]') if stats['# Trades'] else None
                    for stats in (Backtest(df, strategy, **bt_kwargs).run(**run_kwargs)
                                  for df in dfs)]
        finally:
            for shmem in chain(*shms):
                shmem.close()

    def optimize(self, **kwargs) -> pd.DataFrame:
        """
        Wraps `backtesting.backtesting.Backtest.optimize`, but returns `pd.DataFrame` with
        currency indexes in columns.

            heamap: pd.DataFrame = btm.optimize(...)
            from backtesting.plot import plot_heatmaps
            plot_heatmaps(heatmap.mean(axis=1))
        """
        heatmaps = []
        # Simple loop since bt.optimize already does its own multiprocessing
        for df in _tqdm(self._dfs, desc=self.__class__.__name__):
            bt = Backtest(df, self._strategy, **self._bt_kwargs)
            _best_stats, heatmap = bt.optimize(  # type: ignore
                return_heatmap=True, return_optimization=False, **kwargs)
            heatmaps.append(heatmap)
        heatmap = pd.DataFrame(dict(zip(count(), heatmaps)))
        return heatmap


# NOTE: Don't put anything below this __all__ list

__all__ = [getattr(v, '__name__', k)
           for k, v in globals().items()                        # export
           if ((callable(v) and v.__module__ == __name__ or     # callables from this module
                k.isupper()) and                                # or CONSTANTS
               not getattr(v, '__name__', k).startswith('_'))]  # neither marked internal

# NOTE: Don't put anything below here. See above.