How could one actually compute log returns of a series and eventually return sign by making use of an indicator class?

I have tried with the following code, but not great success:

class RET(bt.Indicator):
    '''
    Compute Log Returns
    '''
    # Indicator Variables
    plotlines = dict(RET=dict(_method='bar', alpha=0.50, width=1.0))
    lines = ('RET', 'SIGN')    
    
    def __init__(self):      
        # Store Values to Lines 
        sign = abs(self.data.close - self.data.open) / (self.data.close - self.data.open)
        self.lines.SIGN = sign
        self.lines.RET = math.log(self.data.close) - math.log(self.data.close[-1])

This ends up in the following errors:

<ipython-input-47-c83fca8591c6> in __init__(self)
     12         #sign = (self.data.close - self.data.open) / (self.data.close - self.data.open)
     13         #self.lines.SIGN = sign
---> 14         self.lines.RET = math.log(self.data.close) - math.log(self.data.close[-1])

TypeError: a float is required

or

<ipython-input-50-cf0b74bdd9aa> in __init__(self)
     10     def __init__(self):
     11         # Store Values to Lines
---> 12         sign = (self.data.close - self.data.open) / (self.data.close - self.data.open)
     13         self.lines.SIGN = sign
     14         #self.lines.RET = math.log(self.data.close) - math.log(self.data.close[-1])

TypeError: unsupported operand type(s) for /: 'LinesOperation' and 'LinesOperation'

I suppose this errors are because of the way my code is accessing the data. Any hints on how to perform this calculations?

Thank you!

Hi @backtrader,

To start with, congratulations for the neat framework you have here came up with. Have been fiddling around since few weeks ago with it, and i am surprised with the simplicity of use, yet powerful abilities available.

Going to the subject under discussion, I would like to give it a go and try some implementation of different bars, initially with Heikin-Ashi and later on with Renko or Range bars.

The first question would be if you have any advice on where to plug the code for those bars? After looking into backtrader docs, i wonder if Filters are the right way to do it, after looking at it's description:

'Filters transform the values provided by a data feed to deliver a different data feed' ?
Or rather one shall just previously compute the bars and just add them as a regular datafeed to cerebro? Obviously this would loose later on the live trading ability and just be useful for backtesting purposes.

As for the issue with datetime stamping observations for price based bars, i don't see an issue there, as from my pov they shall have anyhow a time-stamp associated with the time they were started/per observation, therefore comparable with regular bars, just not with a regular interval. Apart from plotting which shall be non sense, the rest should work. Does this make any sense?

Cheers