math.log with linebuffer._LineDelay object

marsario

Hello!

This is my indicator:

class weightedlogmeanInd(bt.Indicator):

    lines = ('mean',)  # output line (array)
    params = (
        ('speed', 4),
        ('tau',0.01),
    )

    def __init__(self):
        weights = findweights(self.p.speed,self.p.tau) #this is an array
        change_sum = 0
        period = len(weights)
        for day in range(0,-period,-1):
            change = self.data.close(0-day) - ( self.data.close(-1-day))
#            change = math.log(self.data.close(0-day)) - math.log( self.data.close(-1-day))
            weighted_change = change * weights[-day]
            change_sum = change_sum + weighted_change 
        self.lines.mean = change_sum / period

I would like to change this line:

change = self.data.close(0-day) - ( self.data.close(-1-day))

into the line below, which is now commented out:

change = math.log(self.data.close(0-day)) - math.log( self.data.close(-1-day))

When I do it, I get the following error message:

TypeError: must be real number, not _LineDelay

Basically, I would like to use the logarithm to get the price change from yesterday to today. What would be the best way to do it?

Thanks a lot!

run-out

@marsario There's a built in Log function in backtrader. It will provide you with the log of a single number in next, not for a whole line in init.

All of this:

for day in range(0,-period,-1):
change = self.data.close(0-day) - ( self.data.close(-1-day))
change = math.log(self.data.close(0-day)) - math.log( self.data.close(-1-day))
weighted_change = change * weights[-day]
change_sum = change_sum + weighted_change

Needs to move into next, and you don't need range.
I haven't tested the following, but generally it should look something like:

def next(self):
    self.close_log[0] = bt.Log(self.datas[0].close[0])
    self.change[0] = self.close_log[0] - self.close_log[-1]
    self.weighted_change[0] = self.change[0] * self.weights[0]
    self.change_sum[0] = self.change_sum[-1] + self.weighted_change[0]

With respect to indexing, you don't have to go back the extra [-1] since backtrader sees the bar at [0] as closed.

marsario

@run-out said in math.log with linebuffer._LineDelay object:

def next(self):
self.close_log[0] = bt.Log(self.datas[0].close[0])
self.change[0] = self.close_log[0] - self.close_log[-1]
self.weighted_change[0] = self.change[0] * self.weights[0]
self.change_sum[0] = self.change_sum[-1] + self.weighted_change[0]

Hi, thanks for your suggestion!

I'm getting AttributeError that backtrader doesn't have a Log method. That's suppose to be an indicator, isn't it?

If so I should be able to recreate it.

marsario

@run-out

I'm trying to implement the indicator with a different logic. And I don't understand what happens.

class weightedlogmeanInd(bt.Indicator):

    lines = ('mean',)  # output line (array)
    params = (
        ('speed', 4),
        ('tau',0.01),
    )
    weights = []

    def __init__(self):
        self.weights = findweights(self.p.speed,self.p.tau)
        today=self.data.get(ago=0,size=len(self.weights))
        yesterday=self.data.get(ago=-1,size=len(self.weights))
        print("len(today):",len(today))
        print("len(yesterday):",len(yesterday))
        print("What is happening in today?", (today) )
        print("What is happening in yesterday?",(yesterday) )
        
        growth = np.log(today)-np.log(yesterday)
        
        print("data close:",self.data.close(0))
        print("What is happening in growth?", growth )

The printout and error message:

(base) Simones-Air:Learning Backtrader simoneromeo$ python 26\ onestock\ strategy.py 
Launching the strategy
Starting logmean strategy
len(today): 0
len(yesterday): 27
What is happening in today? array('d')
What is happening in yesterday? array('d', [133.74, 137.18, 136.76, 136.91, 136.01, 135.39, 135.13, 135.37, 133.19, 130.84, 129.71, 129.87, 126.0, 125.86, 125.35, 120.99, 121.26, 127.79, 125.12, 122.06, 120.13, 121.42, 116.36, 121.09, 119.98, 121.96, 121.03])
Traceback (most recent call last):
  File "26 onestock strategy.py", line 150, in <module>
    cerebro.run()
 ...
  File "26 onestock strategy.py", line 46, in __init__
    growth = np.log(today)-np.log(yesterday)
ValueError: operands could not be broadcast together with shapes (0,) (27,)

I don't undestand: why does self.data.get(ago=0,size=len(self.weights) return an empty array and self.data.get(ago=-1,size=len(self.weights) return an array with the right length?

marsario

@run-out

After some unsuccessful attempts, i came back to the strategy you recommended. I defined log like this:

from backtrader.functions import MultiLogic


class Log(MultiLogic):
    flogic = math.log

Also, added one more line to the code you recommended so that I passed the mean to the line object:


class weightedlogmeanInd(bt.Indicator):

    lines = ('mean',)  # output line (array)
    params = (
        ('speed', 4),
        ('tau',0.01),
    )
    
def next(self):
    self.close_log[0] = Log(self.datas[0].close[0])
    self.change[0] = self.close_log[0] - self.close_log[-1]
    self.weighted_change[0] = self.change[0] * self.weights[0]
    self.change_sum[0] = self.change_sum[-1] + self.weighted_change[0]
    self.lines.mean[0]= bt.Sum(self.change_sum)/len(self.weights)

This doesn't send an error, but doesn't behave as I wished (ideally, it should create a Mean line for each day of the datafeed. I realize that I don't understand how to use Indicators' Next method (I have created indicators but always inside Init). I wish I could find more resources to learn – Platform concepts and Indicators Development don't help me enough apparently. Do you have recommendations of how I could learn to build this indicator?

ab_trader

@marsario said in math.log with linebuffer._LineDelay object:

I'm getting AttributeError that backtrader doesn't have a Log method. That's suppose to be an indicator, isn't it?

i think @run-out used the log function implemented in backtrader2. if you use baseline bt package, than it is not there.

marsario

@ab_trader Is there a way still to find the logarithm inside an indicator? The solution that I thought I found is a mess and it won't work.

run-out

@ab_trader woops!

run-out

@marsario said in math.log with linebuffer._LineDelay object:

self.lines.mean[0]= bt.Sum(self.change_sum)/len(self.weights)

I think you are close. This line:

self.lines.mean[0]= bt.Sum(self.change_sum)/len(self.weights)

The self.change_sum is already a sum from the previous line. It's cummulative. So I think this line should read:

self.lines.mean[0]= self.change_sum[0]/len(self.weights)

Also, in your first line.

self.close_log[0] = math.log(self.datas[0].close[0])

Import math and try that. self.datas[0].close[0] is a scalar and math.log should work.