Not logged in
Viewing 15 posts - 1 through 15 (of 15 total)
  • #262281 |
    Customer
    17 Posts

    Dear Community and Admin,

    First, I like to extend my appreciation to contribution of active community, SQ admin and people who have helped me so far. I have made several mistakes – even by posting a wrong article under the Shared section, without knowing my silly mistake till now. I hope SQX admin don’t bar me later :)

    I am not only new in SQX, but also building EA. I am not a full time trader. With a full time job that take up most ofmy free time. Hence SQX is like my hobby after work around 11pm-2am. I always wanted to learn algo trading and hopefully develop an EA to offload myself from trading, and hopefully replace my full time job. :) However, i am targetting around 5-10% per month.

    The training material by SQX team and a number of discussions here are really amazing. They taught me a lot. I am not able to finish all posting yet. So pardon me if this question had been posted previously.

    I follow the course to develop my first EURUSD H1. While it is started on a demo account. I realise that it is not ideal and i would need to spend more time over many more weekends to retest and re-optimise. But this is weekend time. I don’t want to let the CPU resources to be left idle, so i started a second project on EURUSD M15. There are not much info. I am not sure the following ranking settings make sense. Would appreciate someone to shed some light. I have also attached my builder configuration in case i left something.

    – Profit Factor (IS/OOS) = 1.3

    – Ret/DD Ration (IS/OOS) = 1.5

    – # of trades = 200

    – Winning percentage = 40%

    Grateful to your advise.

     

    Attachments in this forum are visible only to Customers.
    #262306
    Customer
    734 Posts

    not much bad on your settings…

    always use cloned data to your timezone of your broker, where you will be trading, it seems that you are using not cloned data, so the backtest  will not be realiable

    for stop strategies always use slippage in your backtests – 1 pip should be enough

    you are generating M15 with 1M precision? – for builder use selected TF only and for M15 you will need time and strong PC specs, what PC spec you have? And doesnt make sense to generate on 1M and have turned on another crosscheck for higher backtest precision, where you are using the same 1M precision

    by my opinion you are using too many buildling blocks in EXIT TYPES – you have everything selected and thus will lead to a slower computation

    why you are using fixed MM? will you have enough capital to trade it?

    your ranking in nonsense, because of the IS/OOS periods, which are not the same – your IS lasts 11 years, and you OOS is only 4 years – you really want to build on 11 years of data? and ratios which are dependant on time like return/drawdown cant be the same for 11 vs 4 years and number of trades for IS/OOS should be also adapted to your IS/OOS periods – why to use 200 trades for both if the ratio between IS/OOS is 11:4? winning percent 40% is unnecessarily big, 30% could be enough

    You want to be a profitable algotrader? We started using StrateQuant software in early 2016. For now we have a very big knowhow for building EAs for every possible types of markets. We share this knowhow, apps, tools and also all final strategies with real traders. If you want to join us, fill in the FORM. 1500+ final SQX strategies for members running on demo accounts to verify the edge DEMO ACCS. We provide also strategies for indices - DAX, DOW JONES, NASDAQ, SP, UK, because we have realtick data from our brokers.

    #262354
    Customer
    17 Posts

    Dear Hankeys,

    Thank you. I address each your question and may have more to ask. Hope you don’t mind as i want to learn from you to perfect my skill. Note: [H] is you. [G] is me.

    [H] always use cloned data to your timezone of your broker, where you will be trading, it seems that you are using not cloned data, so the backtest  will not be realiable

    [G] I have made the changes. See image. Thank you.

    [H] for stop strategies always use slippage in your backtests – 1 pip should be enough

    [G] I have made the change. In fact, it should be 1 pip. Not sure why. Please see images. Thank you.

    [H] you are generating M15 with 1M precision? – for builder use selected TF only and for M15 you will need time and strong PC specs, what PC spec you have? And doesnt make sense to generate on 1M and have turned on another crosscheck for higher backtest precision, where you are using the same 1M precision

    [G] I have changed “Data” to use “Selected timeframe only”. And keep and “Crosscheck for higher backtest precision” to use “1M data tick”. Am i correct? Please see images. I have 6 CPU cores and 10GB memory now.

    [H] by my opinion you are using too many building blocks in EXIT TYPES – you have everything selected and thus will lead to a slower computation

    [G] It was default. I did not make change since installation. I have made the adjustment now. Please see image. Your advice, if incorrect. Thank you.

    [H] why you are using fixed MM? will you have enough capital to trade it?

    [G] Indeed don’t need fixed $100 MM. I have changed to “Risk fixed % of account”. See setting in image. I intend to fund account USD3k to USD5k on 1:200 leverage account.

    [H] your ranking in nonsense, because of the IS/OOS periods, which are not the same – your IS lasts 11 years, and you OOS is only 4 years – you really want to build on 11 years of data? and ratios which are dependant on time like return/drawdown cant be the same for 11 vs 4 years and number of trades for IS/OOS should be also adapted to your IS/OOS periods – why to use 200 trades for both if the ratio between IS/OOS is 11:4? winning percent 40% is unnecessarily big, 30% could be enough

    [G] Thank you for pointing out. I may have seen and copied from info from material. Actually current IS should be ~13.5 years. Now… i change to the following:

    IS – 7 years (2014-01-01 till 2010-12-31) and OOS – 7 years (2011-01-01 till 2017-12-31).

    Keep to 200 trades for both IS/OOS. Change winning percentage to 30% for both IS/OOS. Maintain Profit Factor as 1.3 for both IS/OOS. Maintain Ret/DD ration for both IS/OOS. See images.

    Lastly, i appreciate your patience. I have attached cfx builder configuration file. Once again, thank you.

     

    Attachments in this forum are visible only to Customers.
    #262358
    Customer
    734 Posts

    your broker is really UTC+8? which one is it?

    i dont like %MM, i am using only fixedsize

    i dont like market strats, i am trading only stop ones

    You want to be a profitable algotrader? We started using StrateQuant software in early 2016. For now we have a very big knowhow for building EAs for every possible types of markets. We share this knowhow, apps, tools and also all final strategies with real traders. If you want to join us, fill in the FORM. 1500+ final SQX strategies for members running on demo accounts to verify the edge DEMO ACCS. We provide also strategies for indices - DAX, DOW JONES, NASDAQ, SP, UK, because we have realtick data from our brokers.

    #262379
    Customer
    17 Posts

    Dear Hankeys,

    Once again, thank you so much.

    [H] your broker is really UTC+8? which one is it?

    [G] I was like dreaming until you punch on my face to tell me  i was wrong on time zone. My broker MarketWatch is showing GM+2, and i confirm there is DST based on a posting in social site. So i have fixed it by choosing a city with DST. Many thanks.

    [H} i dont like %MM, i am using only fixedsize

    [G] For M15, it makes sense to fix at 0.01 lot size indeed.

    [H] i dont like market strats, i am trading only stop ones

    [G] I have adjusted. If we trade STOP, i believe it should include LIMIT order too. Otherwise, it does not make sense. Actually i am not sure if it building block for Exit is good enough.

    Attachments in this forum are visible only to Customers.
    #266748
    Customer
    17 Posts

    Dear Community,

     

    Any kind soul can shed my some light and pull me out from this. :)

    I have a builder running for a short time before the SQ app crashed. I try to change various memory and data setting, but to no availd.

     

    My server is powered by 6 CPU cores and 10Gb memory.  I use the following parameters in my shortcut. I had also tried changing “-J-Xmx9g” to “-J-Xmx13g”. None helps.

    C:\StrategyQuantX\StrategyQuantX.exe -J-server -J-Xmx9g -J-XX:+DisableExplicitGC -J-XX:+AggressiveOpts -J-XX:+UseSerialGC

     

    In the memory config, i used “Let program determines max memory” and also fixed it to “9G”. The result is the same. I am wondering if my Genetic setup in build too resource intensive for a EURUSD M15 trade. Or did i use some silly / wrong config setting.

     

    Appreciate your advise and i thank you in advance…

    Attachments in this forum are visible only to Customers.
    #266754
    Mark Fric
    Administrator
    1177 Posts

    gbjohn – from the message it is clear that SQ tun out of memory, so even 9GB RAM is not enough.

     

    I looked at your config, there is nothing specific, but try to decrease number of Islands to 2 and maybe max. number of strategies in databank to 1000.

    There is something in the config that is causing it to use all the available memory, Java startup configuration will not help it. But I suggest you keep memory fixed to 9GB.

    Mark
    StrategyQuant architect

    #266765
    Customer
    17 Posts

    Dear Mark,

     

    Thank you. Something is not right. I have change from 4 islands to 2 islands. While SQ app does not crash now, and looks running, but it is stuck for last 12-13 hours.

    I am not sure if it is hardware related or was caused by OS patches. I will consider reinstalling the apps.

    Lastly, i would like to know. I have planned to find a new hosting provider (more value of money with more CPU/memory resources and lesser cost), this is probably the third in 3 months since i started this journey. I hope there is no issue with licensing and  other support matter later.

     

    Thank you in advance. Appreciate you all for the support and assistance rendered.

    Attachments in this forum are visible only to Customers.
    #266806
    Customer
    17 Posts

    Dear Community:

    I delete and reinstall SQX, no memory issue anymore. However, each accepted strategy takes too long. I read the algo trading material (chapter 2 to 5) a few times to confirm i understand but i may not. Hence I am sharing the configuration information and Builder config file. Let me know all these make senses:

     

    1. Data

    • IS 6 years from 2004-01-01 to 2009-12-31
    • OOS 6 years from 2010-01-01 to 2015-12-31
    • Spread: 2, Slippage: 1 and Distance: 0
    • Is 6+6 years too much for M15 timeframe? I will keep 2016-01-01 to present for robustness tests later.

    2.  What to build

    • Simple strategy
    • Both, SQX Signal and Genetic.
    • Condition in entry and exit: Min 1 Max 3
    • Global Indicator: Min 5 and Max 200
    • Global Lookback Period: Max 1
    • Stop Loss and Profit Target: Fixed 15 to 100 pips, ATR Multiple 1.5 to 3, ATR Period 10 to 20 (about 2.5 to 5 hours for M15 timeframe). Use indicator is disabled.

    3. Genetic Options

    • 100 Generations, 100 population size per island, Crossover 90%, Mutation 30%
    • 4 islands, migration every 20 generations, 10% population migration rate.
    • Initial population generation and generated decimal coefficient is disabled.
    • Filtered generated initial populated is disabled.
    • Fresh blood detects same strategies is enabled, replace 10% weakness with new one every 2 generations.
    • Evolution management: start again when finished and restart evolution if fitness of in-sample (whole) stagnates for 3 generations

    4. Trading Options

    • Exit on Friday 23:00 (GMT+2 hours)
    • Min StopLoss and TargetProfit: 15, Max StopLoss and TargetProfit: 100

    5. Building Blocks

    • I am not very experience for M15 timeframe. What is/are not important so that I can eliminate it/them. In general, for Forex, I only disable Average Volume and Bar & Time.
    • 128 Signals, 45 Indicators, 29 Stop/Limit entry blocks were selected. Details in attached Builder config. Appreciate some advice to reduce builder processing time, if the selections do no make sense.
    • For order types, since M15 Forex and EURUSD market, I select only STOP order (looking for breakthrough in prices only).
    • For exit types,  default SL/TP, Move SL to BreakEven with 3 pips added. Trailing Stop and Trailing Activation were selected with default. Exit after Bars and ExitRules are disabled.

    6. Money Management

    • Risk $100 for 2 decimal so lot size can be in multiples of 0.01, size if no MM is 0.01 for 10 lot max, so total expected should not be > 0.1 lot.

    7.  Cross Checks

    • 1min precision, 2 spread to be same as Data setup.
    • Filtering, >= 90% net profit, >= 90 # of trader, < 110% max drawdown%.

    8. Ranking and  filtering

    • Auto remove no-trade only.
    • Profit factor for IS and OOS is same, each setting is 1.3.
    • Return / Drawdown ration based on 0.5 per year, for 6 years data,  each IS and OOS setting is 3.
    • Winning percentage for IS and OOS is same, each setting is 30%.
    • For number of trades, i expect about 4 trades per week for M15, or 200 trades per years, for 6 years i asked for > 1000 trades, which i think is reasonable to avoid overfitting.

    I appreciate your help to see if the above is making sense. Details in config file attached. Thank you all in advance…

    • This reply was modified 3 months, 3 weeks ago by gbjohn701.
    • This reply was modified 3 months, 3 weeks ago by gbjohn701.
    Attachments in this forum are visible only to Customers.
    #266809
    Customer
    734 Posts

    problem 1 – exiting friday is for me too late, there will be spread widened

    problem 2 – 1000 trades for this period of time is nonsense, with higher number of trades you will not gain more profitability, only more costs – for me its simple – filter number of trades for example as 20-30 trades average per year

    and its obvious that for M15 you need a strong PC specs, what do you have?

    You want to be a profitable algotrader? We started using StrateQuant software in early 2016. For now we have a very big knowhow for building EAs for every possible types of markets. We share this knowhow, apps, tools and also all final strategies with real traders. If you want to join us, fill in the FORM. 1500+ final SQX strategies for members running on demo accounts to verify the edge DEMO ACCS. We provide also strategies for indices - DAX, DOW JONES, NASDAQ, SP, UK, because we have realtick data from our brokers.

    #266812
    Customer
    17 Posts

    Dear Hankeys,

     

    Thank you for being the first to rescue me. Indeed, i think 1000 trade is not possible, so i reduced to 500, then 200 and 100, still don’t get any returns. So I delete the ranking rules. With no number of trades in the rules, then i saw the trade can be max 41 trade or even < 10 trades for 6 years, It does not make sense. I believe some building blocks that do not make sense could load the processing.

     

    I have 8 core CPUs and 10GB RAM. I am not sure if this is powerful enough.

     

    With these hardwares, does the config make sure, or i have to be more patient?

    Attachments in this forum are visible only to Customers.
    #266822
    Customer
    734 Posts

    8 cores, but what CPU?

    and i dont like much your genetics settings (you are restarting it very soon, why are you using migration, etc.?) and also why are you generating on mostly whole data 2004-2015 and why M15?

    You want to be a profitable algotrader? We started using StrateQuant software in early 2016. For now we have a very big knowhow for building EAs for every possible types of markets. We share this knowhow, apps, tools and also all final strategies with real traders. If you want to join us, fill in the FORM. 1500+ final SQX strategies for members running on demo accounts to verify the edge DEMO ACCS. We provide also strategies for indices - DAX, DOW JONES, NASDAQ, SP, UK, because we have realtick data from our brokers.

    #266823
    Customer
    17 Posts

    Actually it is not a physical CPU, but vCPU powered on VMWare. Xeon CPU E5-2420 @ 1.9GHz. Sorry my bad, 6 instead of 8.

    I have changed to restart evolution after 5 generations instead of 3. I was thinking to reduce unnecessary processing if it stagnates.

    Why in-sample whole data? This is default. I have attended the online algo training, which has only in-sample data on older version of the software, the new version has 3 options – in-sample (whole), in-sample (training) and in-sample (validation). So i left setting as default for whole data. I almost want to change “in-sample (validation)”, but i thought may it is good to apply the whole in-sample data. I have now changed it to “in sample (validation)”. I am a newbie, and i don’t know if there are material that I should read and I have missed. So, I don’t know what i don’t know, and I just use the default to start with.

    Why M15? This is my second project. I started with EURUSD H1 by learning and copying the settings from the course. Then, putting it on a demo account. I started M15 as second project while waiting and checking. I want to change the way to create new EA, so that i can learn outside the training material faster – which were not covered in great details. Between choosing a new market like GBPUSD H1 or USDCHF H1, i had chosen EURUSD M15 instead, which I thought should be “easier” as there is a lot of pound movement due to Brexit, and i am not so familiar with USDCHF. I know M15 is not really easier as the lower time frame means more processing power. It is also difficult to do robustness testing or result is not accurate from real trading. Yes?

    After having two EURUSD H1 strategies on demo, I noticed both are not perfect and ideal, I had made some changes before putting them back on demo in subsequent week. I should go back to make my H1 EA better, but i have started a little too deep on M15. So i decide to continue to force myself to learn. Afterall i started this journey to perfect my skill and learn algo training. To learn, I need to explore outside the material and not just stick to one. I am happy to go back to my EURUSD H1.

    This is my learning journal. I am flexible to adjust in order to learn.

    #266824
    Customer
    734 Posts

    so you are trying to generate on VPS or dedicated server?

    You want to be a profitable algotrader? We started using StrateQuant software in early 2016. For now we have a very big knowhow for building EAs for every possible types of markets. We share this knowhow, apps, tools and also all final strategies with real traders. If you want to join us, fill in the FORM. 1500+ final SQX strategies for members running on demo accounts to verify the edge DEMO ACCS. We provide also strategies for indices - DAX, DOW JONES, NASDAQ, SP, UK, because we have realtick data from our brokers.

    #266965
    Customer
    17 Posts

    Dear Hankeys,

     

    Yes, indeed it is a VPS server now. Even if it is not a dedicated server, do you think based on the builder config, it can take about 10-14 hours to accept a strategy? Or on VPS with 6 vCPU and 10GB RAM this is expected?

Viewing 15 posts - 1 through 15 (of 15 total)

You must be logged in to reply to this topic.