Backtest multitreading
#1
Its possible? Now backtest use only 1 thread. NN strats tests take a lot time. Maybe when supporting other cpu cores will be better.
  Reply
#2
(02-22-2018, 05:21 PM)xFFFFF Wrote: Its possible? Now backtest use only 1 thread. NN strats tests take a lot time. Maybe when supporting other cpu cores will be better.

I would like to join into the request. I am optimizing strategy parameters via python script and every backtest on 2 months data takes about 5 seconds. Thus, with some 100k points to test at least, it takes unreasonably long time. Any multithreading alternative would be greatly appreciated
Thanks
  Reply
#3
(02-25-2018, 07:58 PM)god Wrote: I would like to join into the request. I am optimizing strategy parameters via python script and every backtest on 2 months data takes about 5 seconds. Thus, with some 100k points to test at least, it takes unreasonably long time. Any multithreading alternative would be greatly appreciated
Thanks

Not relevant to the feature request, but if you use something like gekkoga to optimise your parameters you can run tests in parallel. Set as may parallel tests as you have cores Smile
  Reply
#4
(02-27-2018, 11:27 AM)Gryphon Wrote:
(02-25-2018, 07:58 PM)god Wrote: I would like to join into the request. I am optimizing strategy parameters via python script and every backtest on 2 months data takes about 5 seconds. Thus, with some 100k points to test at least, it takes unreasonably long time. Any multithreading alternative would be greatly appreciated
Thanks

Not relevant to the feature request, but if you use something like gekkoga to optimise your parameters you can run tests in parallel. Set as may parallel tests as you have cores Smile

It is not that relevant, you are correct. I did not know about gekkoga, so I wrote my own python script that optimizes parameters by parallelizing hte backtesting onto the given number of cores. However, this still takes quite some time if you want a few million datapoints
  Reply


Forum Jump:


Users browsing this thread: