138% of Retester Complete

Sometimes see above 100% complete in re-tester or optimizer.



Usually after optimizer or re-tester has failed  to 'hook up'...this is random. I have posted before about this in previous versions. It will hang on zero % forever, when you hit stop, it stops and reports how far it had got in the % bar...it just wasn't reporting this as it progressed. 


Few restarts (hitting stop, start a few times) and it hooks up, but will sometimes then complete with a value of over 100%.


All this makes it difficult to trust whats going on under the hood and the results produced.

Attachments
138.jpg
(556.73 KiB)
  • Votes 0
  • Project StrategyQuant X
  • Type Bug
  • Status Fixed
  • Priority Normal

History

N
#1

nathan

27.11.2018 13:00

Task created

MF
#2

Mark Fric

28.11.2018 08:20

Status changed from New to Waiting for information

which of the slow cross checks do you use in that example?


I'll check it, but the problems you describe are most probably caused by starting Retester again when it wasn't fully stopped. 


It has no effect on backtesting reliability itself, every backtest is completely separated and not influenced by another possibly running from previous retest.

N
#3

nathan

29.11.2018 10:33

Attachment M1 Simulated WF Only.jpg added

WF only in the corss check, 1M Simulated, max 1000 tests. 18 strategies in databank. In this example, it was a fresh opening of SQX and then I just hit the retest button.


Stays like this with 0% progress for 6 minutes, but with the processors clearly busy...then 6 minutes in suddenly jumps directly to 100% and complete. Running Time and Estimated Time to completion remain at zero for the entire time.


I think now that its doing its job, its  just that the progress bar and time duration/to completion don't work.

N
#4

nathan

29.11.2018 11:47
Some additional information...during the 6 minutes processor usage remains at 75%, then jumps to 100% and the progress bar begins, and everything then completes in around 30 seconds. Processor usage is at 100% during this 30 seconds and the movement of the progress bar. It is as if the program is stuck in a loop of some sort during this 6  minutes then suddenly figures its way out of it and begins. Looking in detail, this 75% always corresponds to 3 of the 4 Numa nodes (3 of the 4 processors in my case) or 45 of the 60 cores.


In builder this issue is not present, and 100% processors usage is attained and maintained from the the get go.


I also check the optimiser and ran the same walk forward in the optimiser...it worked fine, tests began immediately with 100% processor usage from the get go.


The issue is confined to the Retester using walk forward or optimisation profile/paramteter permutation cross checks.

MF
#5

Mark Fric

17.01.2019 15:14

Status changed from Waiting for information to Fixed

this should be fixed now, we modified the way how Retester processes strategies internally, it should be much more effective.

Votes: 0

Drop files to upload

or

choose files

Max size: 5MB

Not allowed: exe, msi, application, reg, php, js, htaccess, htpasswd, gitignore

...
Wait please