Strategy inheritance

Maxim Korobov

I made 3 simple strategies, now I want to move shared code into BaseStrategy with 3 subclasses.

class BaseStrategy(bt.Strategy):
	def __init__(self):
		self.xz = "test"
		self.data = "test"
...
class RSISimple(BaseStrategy):
	def next(self):
		print(self.xz)

Error appeared:

Traceback (most recent call last):
  File "D:/Projects/trading-bot/main/lab/backtrader/netflix.py", line 74, in <module>
	strategies = back_trader.run()
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\cerebro.py", line 809, in run
	runstrat = self.runstrategies(iterstrat)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\cerebro.py", line 926, in runstrategies
	self._runonce(runstrats)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\cerebro.py", line 1276, in _runonce
	strat._oncepost(dt0)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\strategy.py", line 269, in _oncepost
	self.nextstart()  # only called for the 1st value
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\lineiterator.py", line 324, in nextstart
	self.next()
  File "D:\Projects\trading-bot\main\lab\strategy\rsi_simple.py", line 54, in next
	print(self.xz)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\lineseries.py", line 429, in __getattr__
	return getattr(self.lines, name)
AttributeError: 'Lines_LineSeries_LineIterator_DataAccessor_Strateg' object has no attribute 'xz'

Indeed debugger shows that there is no xz or data attributes in RSISimple instance. Even more

print(self.__class__)

shows MetaStrategy, not my BaseStrategy.

How to inherit strategies right?

backtrader

Your code must be a lot more complicated than the snippet shown above which has been for sure typed manually, because inheritance for strategies is not being intercepted.

The following code works 100% and is just like what you in theory have done above.

class MACrossOver1(bt.Strategy):
    params = (
        # period for the fast Moving Average
        ('fast', 10),
        # period for the slow moving average
        ('slow', 30),
        # moving average to use
        ('_movav', bt.ind.MovAv.SMA)
    )

    def __init__(self):
        sma_fast = self.p._movav(period=self.p.fast)
        sma_slow = self.p._movav(period=self.p.slow)

        self.buysig = bt.ind.CrossOver(sma_fast, sma_slow)
        self.xz = 1

    def next(self):
        if self.position.size:
            if self.buysig < 0:
                self.sell()

        elif self.buysig > 0:
            self.buy()


class MACrossOver(MACrossOver1):
    def next(self):
        if self.xz:
            pass

        super(MACrossOver, self).next()

Maxim Korobov

Seems like the problem appears when two init methods we have - in base and subclassed strategy:

import backtrader as bt


class MACrossOver1(bt.Strategy):
	params = (
		# period for the fast Moving Average
		('fast', 10),
		# period for the slow moving average
		('slow', 30),
		# moving average to use
		('_movav', bt.ind.MovAv.SMA)
	)

	def __init__(self):
		self.xz = 1

	def next(self):
		if self.position.size:
			if self.buysig < 0:
				self.sell()

		elif self.buysig > 0:
			self.buy()


class MACrossOver(MACrossOver1):

	def __init__(self):
		sma_fast = self.p._movav(period=self.p.fast)
		sma_slow = self.p._movav(period=self.p.slow)

		self.buysig = bt.ind.CrossOver(sma_fast, sma_slow)

	def next(self):
		if self.xz:
			pass

		super(MACrossOver, self).next()


if __name__ == '__main__':
	back_trader = bt.Cerebro()
	back_trader.adddata(
		bt.feeds.YahooFinanceData(dataname='NFLX', period='h', reverse=True,
								  fromdate=bt.datetime.datetime(2016, 1, 1), todate=bt.datetime.datetime(2016, 12, 31)
								  )
	)
	back_trader.addstrategy(MACrossOver)
	back_trader.run()

Traceback:

Traceback (most recent call last):
  File "D:/Projects/trading-bot/main/lab/strategy/xz_test.py", line 53, in <module>
	back_trader.run()
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\cerebro.py", line 809, in run
	runstrat = self.runstrategies(iterstrat)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\cerebro.py", line 928, in runstrategies
	self._runonce(runstrats)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\cerebro.py", line 1278, in _runonce
	strat._oncepost(dt0)
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\strategy.py", line 269, in _oncepost
	self.nextstart()  # only called for the 1st value
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\lineiterator.py", line 324, in nextstart
	self.next()
  File "D:/Projects/trading-bot/main/lab/strategy/xz_test.py", line 39, in next
	if self.xz:
  File "C:\Users\Home\AppData\Local\Programs\Python\Python35-32\lib\site-packages\backtrader\lineseries.py", line 429, in __getattr__
	return getattr(self.lines, name)
AttributeError: 'Lines_LineSeries_LineIterator_DataAccessor_Strateg' object has no attribute 'xz'

Maxim Korobov

Sorry, I forgot to init base class:

...
class MACrossOver(MACrossOver1):

	def __init__(self):
		MACrossOver1.__init__(self)
		sma_fast = self.p._movav(period=self.p.fast)
		sma_slow = self.p._movav(period=self.p.slow)

		self.buysig = bt.ind.CrossOver(sma_fast, sma_slow)

...

Same for the first example:

BaseStrategy.__init__(self)

backtrader

It's good to know there is no problem.

The base hierarchy in backtrader is designed to avoid having to call __init__ from the subclasses. That means that subclassing from Strategy doesn't need to call any base class __init__.

Background processing is done by means of controlling the __call__ method of metaclasses.

You may want to have a look at this python package:

If your own hierarchy defines __init__ there is no workaround to avoid calling it from subclasses.

Note: there actually is by scanning the classes and automagically invoking the methods without user intervention, but that would also require user cooperation, because the user must not call it. And it breaks some patterns like when some work has to be done before invoking the base class __init__

Maxim Korobov

Now I have to implement abstract strategy using ABC module:

According to some recepies in Python to define Abstract class we should declare that it's abcmeta like this:

class BaseStrategy(metaclass=ABCMeta):

But what if this class is already subclassed from other class?

class BaseStrategy(bt.Strategy):

Use other options (interfaces, exception in base class) instead?

Note that I tried:

class BaseStrategy(bt.Strategy, metaclass=ABCMeta):

With such traceback:

Traceback (most recent call last):
  File "D:/Projects/trading-bot/main/lab/backtrader/netflix.py", line 13, in <module>
	from main.lab.strategy import RSISimple, RSIBuySell, SMACross, SMA_RSI, HolyGrail, MACD_ADX, BBands
  File "D:\Projects\trading-bot\main\lab\strategy\__init__.py", line 1, in <module>
	from .rsi_simple import *
  File "D:\Projects\trading-bot\main\lab\strategy\rsi_simple.py", line 3, in <module>
	from main.lab.strategy.base_strategy import BaseStrategy
  File "D:\Projects\trading-bot\main\lab\strategy\base_strategy.py", line 6, in <module>
	class BaseStrategy(bt.Strategy, metaclass=ABCMeta):
TypeError: metaclass conflict: the metaclass of a derived class must be a (non-strict) subclass of the metaclasses of all its bases

Possible that bt.Strategy uses abstract classes too.

Asked on SO too: http://stackoverflow.com/questions/41659630/abstract-class-which-is-subclass-of-something-else/41659690#41659690

backtrader

backtrader doesn't use abstract classes but it relies heavily on metaclasses and metaprogramming

You are trying to simply replace the entire metaclass hierarchy and python is complaining you cannot replace an existing metaclass hierarchy suddenly at will.

Maxim Korobov

Roger. Any way to solve it?

backtrader

Don't apply abstract classes.

Maxim Korobov

Hopefully solved from the python side, not backtrader library. See answer in SO about one more stub class.
Overall abstract classes makes me a little bit of pain in python. Maybe sometimes I'll master metaclasses :)
Thanks!