Optimization Speed with Back test very slow

Forums › ProRealTime English forum › ProOrder support › Optimization Speed with Back test very slow

Viewing 15 posts - 1 through 15 (of 15 total)
  • 05/15/2023 at 7:13 AM #214693
    Johann

    Hi PRT Team,

    I have for the first time experienced very slow optimization speed with back tests, since Friday last week. Even if my permutations are limited to 100 options/outcomes. Do anyone else have the same problem at the moment or is there something I can do. Al my hardware and software is exactly the same as before for the last 5 years.

    Thank you in advance.

    05/15/2023 at 8:01 AM #214694
    PeterSt

    Hi Johann,

    Is this V12 or V11.1 ?

    PRT-IG or PRT-IB(KR) ?

    Is this for a “Seconds” timeframe and did you perhaps change recently from larger TimeFrames ?

    Is it the calculation of the backtest (the optimization) or is it the showing of the (Detailed Report) result ?

    I did not notice any change, but have been on PRT-IB and V12 only for the past month or so.

    Regards,
    Peter

    1 user thanked author for this post.
    avatar Johann
    05/15/2023 at 8:39 AM #214696
    Nicolas

    You might experience slowdowns during the weekend, especially when many users are doing optimizations at the same time. Is it still the case today?

    1 user thanked author for this post.
    avatar Johann
    05/15/2023 at 9:08 AM #214702
    Johann

    Hi PeterSt,

    Thanks for responding. I am using 11.1 on IG Markets. It is the report that takes long. Usually on 100 permutations it takes no longer than 5 minutes now it takes 4 times as long on a 100k bars in the 5minute timeframe.

    05/15/2023 at 9:09 AM #214703
    Johann

    Yes even right now, but if no one else has the problem I will try and reboot everything…?

    05/15/2023 at 11:41 AM #214713
    PeterSt
    Johann wrote:

    It is the report that takes long.

    Johann, if that is literal, like the showing what you see below – it can well be about internal “errors” in what you apply. It can even combine with the Optimization where “error” as such is not reported but “error trapped” which takes a relative enormous amount of time (the error may occur at each bar and you let pass 100K of them for one iteration (permutation)).

    • Go back to a strategy (of preferably the same but in older fashion) from a time period all was still fine. Does it work well now again ? then try to find something which you caused since and what the PRT backtesting engine may have difficulty with. One simple example : using a color tag at a graph command which tag does not exist (e.g. BLEU instead of BLUE).
    1 user thanked author for this post.
    avatar Johann
    05/15/2023 at 12:57 PM #214734
    aldtrading

    Does the loading bar takes the same amount of time (or even more) from 100% to the exiting of the loading state than from  from 1% to 100% for anyone else? Is it normal behavior?

    05/15/2023 at 2:08 PM #214739
    GraHal
    aldtrading wrote:

    from 100% to the exiting of the loading state

    Generally only a few seconds, but can be maybe near a minute on some strategies.

    Maybe the longer times are due to many users or throttled back server performance (during maintenance etc).

    Have you tried the whole process – Opti and loading – on a really simple Algo?

    Try this off the ChatGPT Thread, time it at opti then load over 100K bars at 5 min TF on DJI and then I’ll do the same and we compare?

    Just opti these at 10 100 10 each

    ShortTermLength = 9

    LongTermLength = 21

     

    05/15/2023 at 2:13 PM #214740
    GraHal

    I got …

    1.   40 seconds to Optimise
    2.  5 seconds to load
    05/15/2023 at 2:22 PM #214741
    Johann

    Thanks will do and see if it is ok….

    05/15/2023 at 2:46 PM #214746
    GraHal

    I also just did it with 10,000 combos over 100,000 bars on 15 min TF and it took …

    1. 25 mins to Opti
    2. 7 seconds to load
    05/17/2023 at 1:54 PM #214846
    aldtrading

    Generally only a few seconds, but can be maybe near a minute on some strategies.

    Maybe the longer times are due to many users or throttled back server performance (during maintenance etc).

    Have you tried the whole process – Opti and loading – on a really simple Algo?

    Try this off the ChatGPT Thread, time it at opti then load over 100K bars at 5 min TF on DJI and then I’ll do the same and we compare?

    Just opti these at 10 100 10 each

    ShortTermLength = 9

    LongTermLength = 21

    Sorry I didn’t have the time to test this until now, I just did it with DJI 100K bars m5 optimising from 10 to 100, 10by10 ShortTermLength and LongTermLength and it took :

    • 14 secs to optimize
    • 3 secs to load

    provided by :

    • optimize you mean: the time it tries every combinations from the first line added to the array until it reaches 100%
    • load you mean: the time it take from when the loading bar disappear after the 100% to when the infos are displayed on the chart and the performance log is opened
    05/17/2023 at 2:43 PM #214848
    GraHal

    So your optimise is not slow at all, you 14 secs and me 40 secs.

    Yes I did mean as you state above in bullet point 4 and 5.

    05/17/2023 at 3:55 PM #214851
    aldtrading

    Maybe you should try it again since you tried it 2 days ago, but yeah for that script it seems ok, I guess the script I’m trying to backtest is relatively slow then.

    Still, I don’t understand why PRT remains stuck on 100% for what seems pretty much the same time it took it to go from 1% to 100% (1h+) and continues to optimize and add new lines in the array, isn’t it supposed to stop when he reaches 100%?

    05/17/2023 at 4:49 PM #214855
    GraHal
    aldtrading wrote:

    remains stuck on 100%

    We can click on a row at any stage of ongoing optimisation, esp when at 100% … we dont have to wait until the equity curve is drawn automatically.

    aldtrading wrote:

    add new lines in the array, isn’t it supposed to stop when he reaches 100%?

    When opti has been particulary slow, I also have noticed above. but it is rare for these last few lines being added to be better results than those already shown in the Table.

    I guess the opti % has to be to the nearest so when it shows as 100% then it may really be 99.90% and so the few last lines are still being added?

  • Author
    Posts
Viewing 15 posts - 1 through 15 (of 15 total)

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
how do i speed up backtest optimization?
strategy work in back test but not in demo
back testing error as you chack longer periods
bar speed timer?
customizing optimization report (adding variable)
extend the time period for back testing
help test new backtesting software
help with backtest that only goes back a few days
optimization oddities with uneven numbers
optimization of systems not possible

Create your free account now and post your request to benefit from the help of the community
Register or Login