Not logged in
Viewing 14 posts - 1 through 14 (of 14 total)
  • #240329|
    Customer
    76 Posts

    Hi,

    I’m Building in SQX B119 and despite the fact that the time/strategy or what’s the same the Strategies per hour is reasoneably high; 50.000 in my case, SQX can’t find more than 1,5 accepted strategies per hour (not Crosschecking to speed up).

    I have the feeling that SQX is indeed very fast compared with SQ3, BUT the results -> Number of strategies in Databank, are more or less the same.

    SQX opens so much the options to search that it ends nowhere. Probably auto-Calibrating is not useful enough? The standard parameters by default are not tuned? I don’t know but the process to get strategies is as slow as SQ3

    #240378
    tomas262
    Administrator
    1105 Posts

    Hello, you can try to update to v120 to compare the results. Let us know

    #240567
    Customer
    76 Posts

    Hi, tried with B120. It’s as slowly as B119 or even worst!!

    Couldn’t it be the AutoCalibrate feature? Is that Calibration really working well? With so many bugs this is a minefield.

     

    I’ve tried to simplify my Configuratikn & Building Block to maximum but the average strategies per hour is less than 1. B119 gave me 1,5 – 2

    #240568
    Customer
    477 Posts

    Must  be something in Your settings that makes the strategies not show up.  On my old Alienware from 2013 i get 2.5 per hour that passed all MC plus WFO during build. One one of my Xeon server using the same build i get 80 per hour. The shorter data you use as IS the less strategies you find per hour even if generated strategies per hour goes up when using Genetic Evo and MC,WFO during build the difference is 4 fold. How ever using up all data during build is not good and this seems to be proven since final strategies in my workflow building with short data is 2.5% and final strategies building with long data is 1% of strategies passed initial build on same settings.

    Calibrating Indicators  works better in 120.

    #240570
    Customer
    76 Posts

    Wow mabi, I wish I had your stats! I removed all the Crosschecks to speed up the process, so no MC, no WF, no Other Markets,…

    I just have the 15 years historic data, 1 minute data precission and 30% OOS all the rest IS. Regarding the Genetic options; 5 Islands. All this together gives me less than 1 strat per hour :(

    #240571
    Customer
    477 Posts

    I generate in Selective timeframe and use 1 minute as crosscheck.

    #240574
    Customer
    76 Posts

    What do you mean with ‘selective timeframe’? Just a portion of the 15 years of historical data? In this case it’s normal you increase speed generation. And 1 minute data precission as crosscheck? So you generate with the selected timeframe precission. Now I understand you get so many strategies.

    I’m not applying your flow but with SQ3 I had the very same flow as now and I got more strategies! Of course the SQ3 engine was clearly slower but being slower it gave me more strategies. Anyone else has this perception? Don’t know what am I missing here

    #240576
    Customer
    76 Posts

    Also, I see on the statistics that I’m generating to a rythm of 45.000 strategies per hour (I guess that’s good enough). But most of the strategies are rejected by the Automatic Filters – Not Trades is the champion with 45% of rejection. Is this normal?

    Attachments:
    You must be logged in to view attached files.
    #240577
    Customer
    148 Posts

    enric, your info will tell us nothing – the speed of generation is the results of your PC specs and your settings – if ill look at the picture you send here i have many whys?

    only comparable results for you speed of PC is the benchmark in the SQX, what value you have, what PC specs you have?

    you know what you are doing?

    why PF 1.4?

    why to set some filter for initial population?

    why to generate on the whole data?

    what are you complete settings?

    how many building blocks?

    what market, what TF?

    how many filters?

    etc. etc. etc.

    :) [b]You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM[/b]

    #240578
    Customer
    477 Posts

    High precision crosscheck was implemented so that you can get a higher output wile generating. Your out put of passed strategies will be much higher with same result . I mean the one that you already find every hour will still be there most probably together with 4 more  :)

     

    #240676
    Customer
    76 Posts

    Hey! Thanks for answering. I’ve been busy these days but back to Forex :)

    A lot of questions @hankeys. The answer for most of them are because that’s my workflow as you have yours. If you’re questioning my workflow that’s something different. None of us is winning real money here. When I see a recipe that really works I’ll go with no hesitation. Anyway I have a detailed answer for any of your questions, I don’t mind to share if you’re interested.

     

    I know that if I use ‘selected timeframe’ instead of 1 minute data on the Building process, or I choose lower historical data I’ll increase performance but that’s kind of cheating yourself. I mean, if you’re working on selected timeframe all the improvement linked with the Genetic process is not reallistic. You’ve been working to improve something that when you run it later on the real data precission could drastically change. For that reason I preffer 1 minute data precission from the very beginning.

     

    If I’m complaining that SQ is slow, it’s because SQ3 being much slower and using the same workflow built more strategies per hour than SQX. And that’s insane! I was just wondering if Calibration feature could have some influence, because frankly I didn’t bother to adjust the Building block default parameters; RSI, etc.

    Regarding the specs of my PC, don’t see the relevance. Same PC for SQ3 that for SQX

    #240677
    Customer
    148 Posts

    its up to you what you do and how…you dont need to listen anybody

    but if you want co compare in someway old SQ and SQX, i must tell – dont :)

    calibration does nothing – only sets some appropriate levels for some building blocks, nothing more. so calibration cant have any impact

    :) [b]You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM[/b]

    #240682
    Customer
    76 Posts

    Oh no. I want to listen everyone who has something interesting to learn to.

     

    I will only follow empty advices of the kind “don’t need to listen anybody”

    #240699
    Customer
    148 Posts

    put here your config file and i will tell you what i am thinking about it :)

    :) [b]You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM[/b]

Viewing 14 posts - 1 through 14 (of 14 total)

You must be logged in to reply to this topic.