Problem mixing inheritance and recursive indicator

justinxjx

I am having a problem getting the subclassed signal from the Trix example when I generate the signal using the recursive EMA version from the blog.

from __future__ import (absolute_import, division, print_function,
                        unicode_literals)

import backtrader as bt
import backtrader.indicators as btind


class MyTrix(bt.Indicator):

    lines = ('trix',)
    params = (('period', 15),)

    def __init__(self):
        ema1 = btind.EMA(self.data, period=self.p.period)
        ema2 = btind.EMA(ema1, period=self.p.period)
        ema3 = btind.EMA(ema2, period=self.p.period)

        self.lines.trix = 100.0 * (ema3 - ema3(-1)) / ema3(-1)


class MyTrixSignalInherited(MyTrix):

    lines = ('signal',)
    params = (('sigperiod', 30),)

    def __init__(self):
        super(MyTrixSignalInherited, self).__init__()
        self.lines.signal = MyEMA(self.lines.trix, period=self.p.sigperiod)


class MyEMA(bt.indicators.PeriodN):
    params = {'period': 30}  # even if defined, we can redefine the default value
    lines = ('ema',)  # our output line

    def __init__(self):
        self.alpha = 2.0 / (1.0 + self.p.period)  # period -> exp smoothing factor

    def nextstart(self):  # calculate here the seed value
        self.lines.ema[0] = self.data[-1] / self.p.period

    def next(self):
        ema1 = self.lines.ema[-1]  # previous EMA value
        self.lines.ema[0] = ema1 * (1.0 - self.alpha) + self.data[0] * self.alpha

Here is a short snippet of the output and the strategy...

class NoStrategy(bt.Strategy):
    params = (('trixperiod', 20),
              ('analyzer', False),)

    def __init__(self):
        self.trix = MyTrixSignalInherited(self.data, period=self.p.trixperiod)

    def next(self):
        print('Trix: %.2f, Signal: %.2f' % (self.trix.trix[0], self.trix.signal[0]))

Trix: -0.07, Signal: nan
Trix: -0.23, Signal: nan
Trix: -0.37, Signal: nan
Trix: -0.53, Signal: nan
Trix: -0.69, Signal: nan
Trix: -0.83, Signal: nan
Trix: -0.97, Signal: nan
Trix: -1.07, Signal: nan
Trix: -1.12, Signal: nan
Trix: -1.11, Signal: nan

Any suggestions on how to get the signal to get calculated properly?

justinxjx

Maybe another hint may help someone help me figure this out. Within the subclass, using the standard EMA indicator works...

class MyTrixSignalInherited(MyTrix):

    lines = ('signal',)
    params = (('sigperiod', 30),)

    def __init__(self):
        super().__init__()        
        self.lines.signal = btind.EMA(self.lines.trix, period=self.p.sigperiod)

It is only when using the recursive version that the nan values appear. By debugging, I am seeing that even within nextstart, the item "self.data" is full of nan values.

Unfortunately, my true use case involves a similarly developed recursive indicator, so I really need to figure out why it seems to be receiving nan values from the lines output of the parent class.

justinxjx

Further testing shows that if I remove the "next" method from the recursive indicator (rendering it non-recursive), that the indicator correctly receives the data from the parent class. That makes me start to think that this is actually a bug.