https://www.backtrader.com/blog/posts/2015-07-23-multicore-optimization/multicore-optimization/

@Wayne-Filkins-0 pass whatever you want to optimize to the maximize function. So for your case, you could multiply a binary value (profitable/unprofitable) by the percentage of profitable trades and pass that to the gen-opt function

Thank you for this code. I made an indicator based on it which measures the median/mean length of zigzag moves:

from collections import deque

import backtrader as bt
import numpy as np

__all__ = ['ZigZag', 'ZigZagLen', ]


class ZigZagLen(bt.Indicator):
    params = (
        ## ZigZag inputs
        ('retrace', 0.05),  # in percent
        ('minbars', 14),  # number of bars to skip after the trend change
        ##

        # ZigZagLen inputs
        ('full_init', True),  #
        ('period', 6 * 90,),  # 6*90  # positive number for rolling series, -1 for expanding series
        ('output', 'median'),  # median or mean
        ('del_outliers', True),

        ('alpha', 0.05),  # exponential smoothing input
    )

    lines = ('combined', 'bull', 'bear')

    def __init__(self):
        self.zigzag = ZigZag(retrace=self.p.retrace,
                             minbars=self.p.minbars, )

        assert self.p.output in ['median', 'mean']

        if self.p.output == 'median':
            self.func = np.median
        elif self.p.output == 'mean':
            self.func = np.mean

        self.stored_listlen = 0

        if self.p.full_init:
            assert self.p.period > 0, 'when full_init is enabled, we need a minimum period'
            self.addminperiod(self.p.period)

    def next(self):
        zz = self.zigzag
        l = self.l
        llist = {l.combined: ['bull', 'bear'],
                 l.bull    : ['bull'],
                 l.bear    : ['bear'], }

        # # No new values. Copying previous values instead of repeating calculations
        if len(zz.lenlist) < self.stored_listlen:
            for line in llist.keys():
                line[0] = line[-1]
            return

        self.stored_listlen = len(zz.lenlist)

        # Do we have at least something to work with?
        if len(zz.lenlist) >= 8:

            if self.p.del_outliers:
                q1 = np.percentile(zz.lenlist, 25)
                q3 = np.percentile(zz.lenlist, 75)
                iqr = q3 - q1
                upperlim = q3 + 1.5 * iqr
                lowerlim = q1 - 1.5 * iqr
            else:
                upperlim = float('inf')
                lowerlim = 0

            lenlist = list(reversed(zz.lenlist))
            if self.p.period > 0:  # Rolling series

                if sum(lenlist) > self.p.period:
                    # finding index of last value in our moving series
                    cumlst = np.cumsum(lenlist)
                    idx = np.where(cumlst > self.p.period)[0][0]
                    lenlist = lenlist[:idx]

            # Getting the length of current move
            last_len = InfoInt(zz.since_last_pivot)
            last_len.bias = 'bear' if lenlist[0].bias != 'bear' else 'bull'
            lenlist.append(last_len)

            for line, bias in llist.items():
                filter_lst = [v for v in lenlist if lowerlim < v < upperlim
                              and v.bias in bias]

                output = self.func(filter_lst)
                if np.isnan(line[-1]):
                    line[0] = output
                else:
                    alpha = self.p.alpha
                    line[0] = line[-1] * (1 - alpha) + output * alpha


# inheriting allows us to add variables to an int
class InfoInt(int):
    pass


class ZigZag(bt.Indicator):
    '''
      ZigZag indicator.
    '''
    lines = (
        'trend',
        'last_high',
        'last_low',
        'zigzag',
    )

    plotinfo = dict(
        subplot=False,
        plotlinelabels=True, plotlinevalues=True, plotvaluetags=True,
    )

    plotlines = dict(
        trend=dict(_plotskip=True),
        last_high=dict(color='green', ls='-', _plotskip=True),
        last_low=dict(color='black', ls='-', _plotskip=True),
        zigzag=dict(_name='zigzag', color='blue', ls='-', _skipnan=True),
    )

    params = (
        ('retrace', 0.05),  # in percent
        ('minbars', 14),  # number of bars to skip after the trend change
    )

    def __init__(self):
        super(ZigZag, self).__init__()

        self.addminperiod(self.p.minbars)

        assert self.p.retrace > 0, 'Retracement should be above zero.'
        assert self.p.minbars >= 0, 'Minimal bars should be >= zero.'

        self.retrace_thresh = self.data.close * self.p.retrace / 100
        self.minbars = self.p.minbars
        self.count_bars = 0
        self.last_pivots = deque([0, 0], maxlen=2)
        self.last_pivot_t = 0
        self.since_last_pivot = 0
        self.lenlist = []

        self.stored_datalen = 0

    def prenext(self):
        self.l.trend[0] = 0
        self.l.last_high[0] = self.data.high[0]
        self.l.last_low[0] = self.data.low[0]
        self.l.zigzag[0] = (self.data.high[0] + self.data.low[0]) / 2

    def next(self):

        # No new candle. Got called due to resampling
        if len(self.data) == self.stored_datalen:
            return
        self.stored_datalen = len(self.data)

        curr_idx = len(self.data)
        self.retrace_thresh = self.data.close[0] * self.p.retrace / 100
        self.since_last_pivot = curr_idx - self.last_pivot_t
        self.l.trend[0] = self.l.trend[-1]
        self.l.last_high[0] = self.l.last_high[-1]
        self.l.last_low[0] = self.l.last_low[-1]
        self.l.zigzag[0] = float('nan')

        # Initialize trend
        if self.l.trend[-1] == 0:
            # If current candle has higher low and higher high
            if self.l.last_low[0] < self.data.low[0] and self.l.last_high[0] < self.data.high[0]:
                self.l.trend[0] = 1
                self.l.last_high[0] = self.data.high[0]
                self.last_pivot_t = curr_idx
            # If current candle has lower low and lower high
            elif self.l.last_low[0] > self.data.low[0] and self.l.last_high[0] > self.data.high[0]:
                self.l.trend[0] = -1
                self.l.last_low[0] = self.data.low[0]
                self.last_pivot_t = curr_idx

        # Up trend
        elif self.l.trend[-1] == 1:
            # if higher high, move pivot location to current high
            if self.data.high[0] > self.l.last_high[-1]:
                self.l.last_high[0] = self.data.high[0]
                self.count_bars = self.minbars
                self.last_pivot_t = curr_idx

            # elif at least p.minbars since last bull swing and currently in a retrace -> Switch Bearish
            elif self.count_bars <= 0 and self.l.last_high[0] - self.data.low[0] > self.retrace_thresh \
                    and self.data.high[0] < self.l.last_high[0]:
                self.switch_to_bear(curr_idx)

            # elif bearish close
            elif self.count_bars < self.minbars and self.data.close[0] < self.l.last_low[0]:
                self.switch_to_bear(curr_idx)

        # Down trend
        elif self.l.trend[-1] == -1:
            # if lower low, move pivot location to current low
            if self.data.low[0] < self.l.last_low[-1]:
                self.l.last_low[0] = self.data.low[0]
                self.count_bars = self.minbars
                self.last_pivot_t = curr_idx

            # elif we had an established bear swing and currently in a retrace -> Switch Bullish
            elif self.count_bars <= 0 and self.data.high[0] - self.l.last_low[0] > self.retrace_thresh and \
                    self.data.low[0] > self.l.last_low[0]:
                self.switch_to_bull(curr_idx)
            # elif bullish close
            elif self.count_bars < self.minbars and self.data.close[0] > self.l.last_high[-1]:
                self.switch_to_bull(curr_idx)

        # Decrease minbars counter
        self.count_bars -= 1

    def switch_to_bear(self, idx):
        self.l.trend[0] = -1
        self.count_bars = self.minbars
        self.l.last_low[0] = self.data.low[0]
        self.l.zigzag[-self.since_last_pivot] = self.l.last_high[0]
        self.last_pivot_t = idx

        self.record_idx(idx, 'bull', )

    def switch_to_bull(self, idx):
        self.l.trend[0] = 1
        self.count_bars = self.minbars
        self.l.last_high[0] = self.data.high[0]
        self.l.zigzag[-self.since_last_pivot] = self.l.last_low[0]
        self.last_pivot_t = idx

        self.record_idx(idx, 'bear', )

    def record_idx(self, idx, bias, ):
        self.last_pivots.append(idx)
        last_move_len = self.last_pivots[-1] - self.last_pivots[0]
        last_move_len = InfoInt(last_move_len)
        last_move_len.bias = bias

        self.lenlist.append(last_move_len)

@Adham-Suliman said in Order.StopTrail turns positive shares into short position:

   elif self.position.size > 0:          
               self.close(exectype=bt.Order.StopTrail, trailamount = 8) 

You're creating a stop order for every next call where your position is open. Assign this order to a variable and only create a new one if it doesn't exist.

@Yulem said in Final SELL order not created in Quickstart Examples:

but no SELL order is generated upon exit leaving me with an open position. I believe the logic is to generate a SELL using the final day's CLOSE value which makes sense and is what I saw in the previous examples.

There is no problem with your code. Whether you still have a position open or closed at the end of a backtest is inconsequential. If you want to generate a sell specifically using the final day's close value, you can do so in the strategy's stop method

@김도언 said in How to make All-in Market Price in tick-chart:

    if self.buy_continue:
        if self.broker.get_cash() > 0:
            self.order = self.buy(size=self.market_volume[0])

market volume will be more than your available cash during your last fill.

a much easier alternative would be something like this inside the notify_order method

if order.status in [order.Margin, order.Rejected]:
    self.order = self.buy(size=int(self.data.close[0] / self.broker.get_cash())

possibly lower size by 1 until you get your fill.

@quake004

for data in datas:
    cerebro = bt.Cerebro()
    cerebro.adddata(data,)

inside this loop set-up everything for backtesting and receive back the results.

@quake004 said in Execution gives many 'Canceled/Margin/Rejected' orders:

    if self.crossover > 0:
        self.log(f'BUY CREATE {self.dataclose[0]:.2f}')
        self.order = self.buy()
    elif self.crossover < 0:
        self.log(f'SELL CREATE {self.dataclose[0]:.2f}')

Did you forget to copy the last line where the sell order is created or not? Other than that, I can't quickly spot something wrong with your code.

@vladisld Had the same issue. problem is that arange returns numpy datatypes which causes this operation to return a numpy array:

stddev = self.p.devfactor * StdDev(self.data, ma, period=self.p.period,
                                           movav=self.p.movav)

Converting self.p.devfactor to a standard float fixes the problem

The guy in your link didn't include fees & slippage.

Further more it should be noted that every single EMA combination tested (and most SMAs) outperformed the buy and hold annualized return of 6.32%^ during the test period (before allowing for transaction costs and slippage)Further more it should be noted that every single EMA combination tested (and most SMAs) outperformed the buy and hold annualized return of 6.32%^ during the test period (before allowing for transaction costs and slippage)

downgrade matplotlib to version 3.2.2

Thank you for this code. I made an indicator based on it which measures the median/mean length of zigzag moves:

from collections import deque

import backtrader as bt
import numpy as np

__all__ = ['ZigZag', 'ZigZagLen', ]


class ZigZagLen(bt.Indicator):
    params = (
        ## ZigZag inputs
        ('retrace', 0.05),  # in percent
        ('minbars', 14),  # number of bars to skip after the trend change
        ##

        # ZigZagLen inputs
        ('full_init', True),  #
        ('period', 6 * 90,),  # 6*90  # positive number for rolling series, -1 for expanding series
        ('output', 'median'),  # median or mean
        ('del_outliers', True),

        ('alpha', 0.05),  # exponential smoothing input
    )

    lines = ('combined', 'bull', 'bear')

    def __init__(self):
        self.zigzag = ZigZag(retrace=self.p.retrace,
                             minbars=self.p.minbars, )

        assert self.p.output in ['median', 'mean']

        if self.p.output == 'median':
            self.func = np.median
        elif self.p.output == 'mean':
            self.func = np.mean

        self.stored_listlen = 0

        if self.p.full_init:
            assert self.p.period > 0, 'when full_init is enabled, we need a minimum period'
            self.addminperiod(self.p.period)

    def next(self):
        zz = self.zigzag
        l = self.l
        llist = {l.combined: ['bull', 'bear'],
                 l.bull    : ['bull'],
                 l.bear    : ['bear'], }

        # # No new values. Copying previous values instead of repeating calculations
        if len(zz.lenlist) < self.stored_listlen:
            for line in llist.keys():
                line[0] = line[-1]
            return

        self.stored_listlen = len(zz.lenlist)

        # Do we have at least something to work with?
        if len(zz.lenlist) >= 8:

            if self.p.del_outliers:
                q1 = np.percentile(zz.lenlist, 25)
                q3 = np.percentile(zz.lenlist, 75)
                iqr = q3 - q1
                upperlim = q3 + 1.5 * iqr
                lowerlim = q1 - 1.5 * iqr
            else:
                upperlim = float('inf')
                lowerlim = 0

            lenlist = list(reversed(zz.lenlist))
            if self.p.period > 0:  # Rolling series

                if sum(lenlist) > self.p.period:
                    # finding index of last value in our moving series
                    cumlst = np.cumsum(lenlist)
                    idx = np.where(cumlst > self.p.period)[0][0]
                    lenlist = lenlist[:idx]

            # Getting the length of current move
            last_len = InfoInt(zz.since_last_pivot)
            last_len.bias = 'bear' if lenlist[0].bias != 'bear' else 'bull'
            lenlist.append(last_len)

            for line, bias in llist.items():
                filter_lst = [v for v in lenlist if lowerlim < v < upperlim
                              and v.bias in bias]

                output = self.func(filter_lst)
                if np.isnan(line[-1]):
                    line[0] = output
                else:
                    alpha = self.p.alpha
                    line[0] = line[-1] * (1 - alpha) + output * alpha


# inheriting allows us to add variables to an int
class InfoInt(int):
    pass


class ZigZag(bt.Indicator):
    '''
      ZigZag indicator.
    '''
    lines = (
        'trend',
        'last_high',
        'last_low',
        'zigzag',
    )

    plotinfo = dict(
        subplot=False,
        plotlinelabels=True, plotlinevalues=True, plotvaluetags=True,
    )

    plotlines = dict(
        trend=dict(_plotskip=True),
        last_high=dict(color='green', ls='-', _plotskip=True),
        last_low=dict(color='black', ls='-', _plotskip=True),
        zigzag=dict(_name='zigzag', color='blue', ls='-', _skipnan=True),
    )

    params = (
        ('retrace', 0.05),  # in percent
        ('minbars', 14),  # number of bars to skip after the trend change
    )

    def __init__(self):
        super(ZigZag, self).__init__()

        self.addminperiod(self.p.minbars)

        assert self.p.retrace > 0, 'Retracement should be above zero.'
        assert self.p.minbars >= 0, 'Minimal bars should be >= zero.'

        self.retrace_thresh = self.data.close * self.p.retrace / 100
        self.minbars = self.p.minbars
        self.count_bars = 0
        self.last_pivots = deque([0, 0], maxlen=2)
        self.last_pivot_t = 0
        self.since_last_pivot = 0
        self.lenlist = []

        self.stored_datalen = 0

    def prenext(self):
        self.l.trend[0] = 0
        self.l.last_high[0] = self.data.high[0]
        self.l.last_low[0] = self.data.low[0]
        self.l.zigzag[0] = (self.data.high[0] + self.data.low[0]) / 2

    def next(self):

        # No new candle. Got called due to resampling
        if len(self.data) == self.stored_datalen:
            return
        self.stored_datalen = len(self.data)

        curr_idx = len(self.data)
        self.retrace_thresh = self.data.close[0] * self.p.retrace / 100
        self.since_last_pivot = curr_idx - self.last_pivot_t
        self.l.trend[0] = self.l.trend[-1]
        self.l.last_high[0] = self.l.last_high[-1]
        self.l.last_low[0] = self.l.last_low[-1]
        self.l.zigzag[0] = float('nan')

        # Initialize trend
        if self.l.trend[-1] == 0:
            # If current candle has higher low and higher high
            if self.l.last_low[0] < self.data.low[0] and self.l.last_high[0] < self.data.high[0]:
                self.l.trend[0] = 1
                self.l.last_high[0] = self.data.high[0]
                self.last_pivot_t = curr_idx
            # If current candle has lower low and lower high
            elif self.l.last_low[0] > self.data.low[0] and self.l.last_high[0] > self.data.high[0]:
                self.l.trend[0] = -1
                self.l.last_low[0] = self.data.low[0]
                self.last_pivot_t = curr_idx

        # Up trend
        elif self.l.trend[-1] == 1:
            # if higher high, move pivot location to current high
            if self.data.high[0] > self.l.last_high[-1]:
                self.l.last_high[0] = self.data.high[0]
                self.count_bars = self.minbars
                self.last_pivot_t = curr_idx

            # elif at least p.minbars since last bull swing and currently in a retrace -> Switch Bearish
            elif self.count_bars <= 0 and self.l.last_high[0] - self.data.low[0] > self.retrace_thresh \
                    and self.data.high[0] < self.l.last_high[0]:
                self.switch_to_bear(curr_idx)

            # elif bearish close
            elif self.count_bars < self.minbars and self.data.close[0] < self.l.last_low[0]:
                self.switch_to_bear(curr_idx)

        # Down trend
        elif self.l.trend[-1] == -1:
            # if lower low, move pivot location to current low
            if self.data.low[0] < self.l.last_low[-1]:
                self.l.last_low[0] = self.data.low[0]
                self.count_bars = self.minbars
                self.last_pivot_t = curr_idx

            # elif we had an established bear swing and currently in a retrace -> Switch Bullish
            elif self.count_bars <= 0 and self.data.high[0] - self.l.last_low[0] > self.retrace_thresh and \
                    self.data.low[0] > self.l.last_low[0]:
                self.switch_to_bull(curr_idx)
            # elif bullish close
            elif self.count_bars < self.minbars and self.data.close[0] > self.l.last_high[-1]:
                self.switch_to_bull(curr_idx)

        # Decrease minbars counter
        self.count_bars -= 1

    def switch_to_bear(self, idx):
        self.l.trend[0] = -1
        self.count_bars = self.minbars
        self.l.last_low[0] = self.data.low[0]
        self.l.zigzag[-self.since_last_pivot] = self.l.last_high[0]
        self.last_pivot_t = idx

        self.record_idx(idx, 'bull', )

    def switch_to_bull(self, idx):
        self.l.trend[0] = 1
        self.count_bars = self.minbars
        self.l.last_high[0] = self.data.high[0]
        self.l.zigzag[-self.since_last_pivot] = self.l.last_low[0]
        self.last_pivot_t = idx

        self.record_idx(idx, 'bear', )

    def record_idx(self, idx, bias, ):
        self.last_pivots.append(idx)
        last_move_len = self.last_pivots[-1] - self.last_pivots[0]
        last_move_len = InfoInt(last_move_len)
        last_move_len.bias = bias

        self.lenlist.append(last_move_len)

Add commission info before running cerebro. Not inside of the strategy.

https://www.backtrader.com/blog/posts/2015-07-23-multicore-optimization/multicore-optimization/

in the run method from cerebro you can intercept the variable product and choose to run only a certain number of optimizations.

lststrats = list(itertools.product(*self.strats))
random.shuffle(lststrats)
lststrats = lststrats[:100]
iterstrats = iter(lststrats)

Using 5 years of hourly data I can run roughly 80-160 sessions at a time. (depending on how much memory observers/indicators take up)

Alternatively you could roughly line up your tests with that number in mind: test in batches of 3 parameters with 6-8 variables for each parameter.

fyi the pathos module has a multiprocessing implementation which uses dill instead of pickle. This solves a lot of pickling errors.

just use self.position.price.

self.positions[data].price if using multiple data sets.

@zuj

You could for each variable, average the return for each parameter and then select the best performing parameter out of that selection

e.g.:

Alternatively, you could include the neighboring parameters as part of the selection process. (fitness value of period=10 is the average of the fitness values of period 8-12)

Many possible solutions.

Export your backtesting data to excel and analyze if over there

Having all results lined up makes it plain as day if a parameter combination is the result of dumb luck or alpha

https://i.imgur.com/N7ZTd8f.png

Would need to see your data. Maybe try removing fromdate & todate arguments. That will use all available data by default