@vladisld, Can you please help me with this issue. This is becoming a deal breaker as Indicator values are key to any automated trading systems. Do you use IB data or some other data source with backtrader? Did you notice any discrepancies with indicator values?
@ab_trader Ok so everything is working now. So the reason for the errors was that in the init for my strategy I initialized the self.gamma variable with gamma1(self.data).dummyline. That was my mistake I should have deleted that .dummyline.
I think I go confused for a bit and accidentally put that there in the init of my strategy. I also used self.line.dummyline in the next() of my strat. which should be used in my indicator class not strat class.
Bunch of dumb mistakes I made. HUGE thanks to @ab_trader and @run-out for helping me out, and clarifying many aspects of BT for me.
@run-out I originally tried this but got a, linebuffer has no attribute dummyline, in the next() function. Shouldn't setting self.gamma to dummyline in the init be enough to call it later on without dummyline
it is needed to be an numpy-array finally, to be understood by Tensorflow.
But anyway, this works, as this is an oneColumn-array with only close-data.
The question is, how to get an OHLC-fourColumn-array, in which the first row  allways actuell ...
I have printed it out and "self.get.data" only gives close-data, not the whole row of data, with OHLC, Volume, date...right?!
I think I have found the solution. I had to use addminperiod function to set minimum period for my Indicator and I had to calculate close slice in next() method and not the __init__ method. I had to assign float value to  of my line and the line itself. Here is the code:
lines = ('radf',)
params = dict(period=600, adf_lag=2)