Wrong value of Indicator in another CustomIndicator

[cgi1]

class SomeStrategy(bt.Strategy):

    def __init__(self):
        self.atr = bt.indicators.ATR(self.datas[0], plotname="ATR", subplot=True)
        self.sr = SR(logger=self.indi_logger, master=self, fromdate=self.p.fromdate_trading, todate=self.p.todate_trading)

Within my custom indicator SR I need to access the ATR line value within SR->next(). So I pass the whole self to it during init:

class SR(bt.Indicator):

    def __init__(self, logger, master, fromdate, todate):
        self.logger = logger
        self.master = master
        self.atr = self.master.atr

In this example I would assume to access the ATR line value of candle 2001-09-10 using self.atr[0] (which is about 0.51). Instead I get the final ATR line value of today (which is 2.67588x):

Do I need to calculate the shift manually to get the correct ATR value or is there another way to achieve this?

Thanks & BR!

Doc Reference

[cgi1]

Generating it new in the custom indicator worked for me:

class SomeStrategy(bt.Strategy):

    def __init__(self):
        self.atr = bt.indicators.ATR(self.datas[0], subplot=False)

[backtrader]

Of course, creating an indicator within an indicator allows the created indicator to report itself to the environment it is being created in. i.e.: the other indicator.

In any case the use case presented in the 1st post can work, but only if you let the platform know (and it doesn't take much)

@cgi1 said in Wrong value of Indicator in another CustomIndicator:

self.atr = self.master.atr

The problem there is: self.master.atr can be anything. It could be a callable. The indicator SR has no way of knowing what's in the attribute atr.

@cgi1 said in Wrong value of Indicator in another CustomIndicator:

In this example I would assume to access the ATR line value of candle 2001-09-10 using self.atr[0] (which is about 0.51). Instead I get the final ATR line value of today (which is 2.67588x)

That's a side effect of the above. Knowing nothing about atr, the indicator SR accesses [0] and because you are running with runonce=True, you get the last calculated value. If SR had known something about atr it would have managed it to give you the right value of [0] each time.

Even if you now think that the problem goes away by doing runonce=False ... yes it would but for the wrong reasons.

What you have to take into account is that mostly everything behaves like a data feed (See the section Almost everything is a Data Feed under Docs - Platform Concepts)

And that's why the proper approach that will work under all conditions is:

class SomeStrategy(bt.Strategy):

    def __init__(self):
        self.atr = bt.indicators.ATR(self.datas[0], plotname="ATR", subplot=True)
        self.sr = SR(self.atr, logger=self.indi_logger, master=self, fromdate=self.p.fromdate_trading, todate=self.p.todate_trading)

followed by

class SR(bt.Indicator):

    def __init__(self, logger, master, fromdate, todate):
        self.logger = logger
        self.master = master
        # NO LONGER NEEDED - atr is `self.data` aka `self.data0` aka `self.datas[0]`
        # self.atr = self.master.atr

Because atr is passed as a data feed to SR it will properly managed under all circumstances.

[cgi1]

Thanks for the good explanation! That´s a view of data feeds I did not had in before.

So self.datas[0] within SR(bt.Indicator) are now set correctly to the values of ATR.

The limitation now is, that there is no YahooDataFeed available anymore, because the data feed gets overwritten by

self.sr = SR(self.atr, [...])

Is it possible to pass multiple data feeds to the indicator?
In this case the market data (YahooDataFeed) and another indicator (ATR)

Something like:

self.sr = SR([self.datas[0], self.atr, self.further_indi], [...])

[backtrader]

Of course. You can pass any number of data-feed-like objects to indicators. They will be available as self.dataX or the corresponding self.datas[X]

In the previous example and because no data feed was explicitly passed, the framework gives the indicator the same data feeds which are available in the strategy (or parent indicator)

[cgi1]

@backtrader said in Wrong value of Indicator in another CustomIndicator:

Of course. You can pass any number of data-feed-like objects to indicators. They will be available as self.dataX or the corresponding self.datas[X]

Nice. For reference:

self.sr = SR(self.datas[0], self.atr)

Gives

inside custom indicator SR. So it can be used like

class SR(bt.Indicator):
    def __init__(self, logger, master, fromdate, todate):

        # self.datas[0] is yahoo market data (as default)
        self.atr = self.datas[1]

DONE