Machine Learning in ProOrder ProRealTime

• 04/22/2020 at 2:15 PM #127459
  Bard

  If you have a huge stoploss, you are more  likely to make money (with luck) even after a big drawdown

  And bear in mind that if you have a 50% DD it will take a massive 100% return for you just to get back to your breakeven!

  1 user thanked author for this post.

  fifi743

  

  Bard

  Participant

  Topics: 138

  Replies: 814

  Been thanked: 99 times
  04/22/2020 at 2:37 PM #127460
  GraHal

  Bard wrote:

  I couldn’t access your VRT?

  You couldn’t access it? Or you couldn’t edit it?

  It should not need to be edited? Open and do as you wish to your copy … maybe you would have to make a copy and rename?? It’s a while since I did a VRT.  So much to do all the time … not even getting much sunshine!

  Vonasi will have to come up for air soon … unless he refreshes via the periscope!!!??? 🙂 🙂

  

  GraHal

  Participant

  Topics: 211

  Replies: 8760

  Been thanked: 2176 times
  04/22/2020 at 2:39 PM #127461
  fifi743

  I couldn’t agree with you more.
  That’s why you need a small SL and a tp adjust has a resistance and that
  The sl – traderprice <traderprice – tp

  2 users thanked author for this post.

  Paul, Bard

  

  fifi743

  Participant

  Topics: 10

  Replies: 902

  Been thanked: 418 times
  04/22/2020 at 2:46 PM #127462
  nonetheless

  I hate to burst the bubble chaps but the robustness score is dismal. You can’t run it with any kind of MM so I changed to N=1. This is on Coffee Arrabica $3.75 Daily (commodities).

  Attached is the itf that I used, if there’s a better version I’m happy to re-run it

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
  04/22/2020 at 3:00 PM #127465
  nonetheless

  Hold the front page, i used the wrong version … robustness is not too bad, ignore previous post!

  2 users thanked author for this post.

  Bard, GraHal

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
  04/22/2020 at 3:20 PM #127469
  nonetheless

  Unfortunately you’ve got a different problem; presently the spread is 60, 3 year results attached. N=1

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
  04/22/2020 at 3:20 PM #127471
  Paul

  The version @nonetheless posted will not work live I think.

  You give a stop order at the renkomax+boxsize, when the high > renkomax+box.

  It will often be rejected live because of minimum stop distance.

  That’s why sometimes results are great  in the backtest but in real they are not  or the strategy is stopped. A market order I think is preferred but less good.  It could perhaps work to run a backtest  with stop on i.e. 1h frame, and use live on 1m like in barhunter with “mode” and market orders.

  secondly,  there is used once to define renkomax/min at the beginning.

  Loading a new chart will have a different first bar and can perhaps influence the backtest, that’s maybe another reason why live & backtest are out of sync.

  2 users thanked author for this post.

  Bard, GraHal

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  04/22/2020 at 4:44 PM #127483
  Bard

  Thanks for your itf, that’s interesting because I ran your itf  “as is” but with a 60 spread and got a 1.46% drawdown and these results?: Screen #1
  Is it a tick by tick test (even if no warning comes up)? I also did it for the 4 hour and got this: Screen #2
  You also have a different if statement for the ling/short conditions including shortonmarket for the long and vica versa for the short.

  I then tried turning off your tradeon to see it test without that code as Vonasi had mentioned not to test money management code with VRT — yet it still did an optimisation?

  Turn of Tradeon

  tradeon = 0 //0 if opendate >= startdate then if barindex mod qty = 0 or barindex mod qty = j then tradeon = 0 //1 endif endif

  1 2 3 4 5 6

  tradeon = 0 //0 if opendate >= startdate then if barindex mod qty = 0 or barindex mod qty = j then tradeon = 0 //1 endif endif

  and then got screen #3. Why the different results? I then (just to make sure!) rem’d them out and got results in screen #4.

  Maybe we should assume it’s also perhaps not tbt testing until I hear back from IG/PRT? Also that 60 spread is a massive… I wonder if that is a normal spread or due to adverse conditions?

  Edit: Solution? I tried a bigger boxsize of 150 (and no VRT) and that works well! £9.3k | Win 46% |Gainloss 1.61% | DD 12%.

  

  Bard

  Participant

  Topics: 138

  Replies: 814

  Been thanked: 99 times
  04/22/2020 at 4:54 PM #127493
  nonetheless

  I was working with this code: #127268

  I only added the VRT and turned off the MM (N =1)

  If you’ve got an update on that, post the itf specifying the instrument and TF … just to make sure we’re not comparing apples and kerosene.

  1 user thanked author for this post.

  bertrandpinoy

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
  04/22/2020 at 5:09 PM #127500
  Bard

  Reinstating the capital/close money mgt algo (not to be used with VRT) also increases expectancy and profits. 14 day range from present and MM ramped way back up to x30 and that’s 6k with less than 1% Ddown IF this is truly tbt testing. Are these results obtainable Live? I know a few have posted good live results from the Renko forum and thanks for that, it would be great to keep seeing those results going forward @eckaw.
  See here: https://www.prorealcode.com/topic/discussion-re-pure-renko-strategy/page/17/#post-126652

  

  Bard

  Participant

  Topics: 138

  Replies: 814

  Been thanked: 99 times
  04/23/2020 at 2:08 PM #127605
  Bard

  Hi Nonetheless, we’re definitely comparing with the system I posted, although there is a slight variation in the If C1 and If C2, as mentioned. (Not sure where that <span class=”difference”>shortonmarket etc </span>came from? It didn’t appear to make any difference in the profits if I replaced my system C1/C’2’s with your “If c1 and c2” statements. It warns that tick by tick (tbt) testing cannot be done too, despite the yellow triangle saying that tests after after the 1st Aug 2010 will be done tbt. Still waiting for IG’s response to that.

  With the “same” code (although yours has VRT) it gives different results and different stop loss GraphX’s. Pls see results image #1. All other things being equal, it is the VRT code effecting the results because when the VRT is rem’d out our results are identical: Pls see results image #2. Wondered if you might know why that is @Vonasi?

  The interesting thing about this comparison test between our systems is that when I ran the identical dates: 24/04/17-22/04/20, as your posted results screen above, with the new 60 spread, your system did not show a warning that tbt testing couldn’t be done. Mine did, and the results are not the same. My code also takes about 4 to 5x’s longer to display results…

  The code is now identical (I’ve checked it 3x’s in the text-compare.com comparison tool). There is no VRT, the dates are the same, the spread is the same, the boxsize is the same and the results were different… the only way I “fixed it” was to change the TF to 4 hours, then I got identical test results, and then I went back to Daily, and then finally the results were the same…Umm. I’ve deleted my slow 1M ML1 StpLoss system and recoded it.

  
  @Nicolas, I wonder if you would know why the tick by tick (tbt) warning didn’t come up for Nonetheless’s version but did for my version when they have the same system code and parameters? I’ve had this tbt warning problem repeatedly, where sometimes it does and sometimes it doesn’t appear, even though I’m testing the same date ranges? Why would it now suddenly not be able to do a tbt test if all I’ve done is change a boxsize from 100 to 50? I also would be interested if you know or could find out how far back the tbt data actually goes for coffee (KC)? Thanks very much.

  The possible solution to that low gain/loss and low profit above is to either use different date ranges (I’ve not created a daily system that lasted longer than 12 yrs) or increase the boxsize from 100 to 150 as mentioned for the daily, (I hadn’t seen your “I was working with this code..” comment when I posted). Pls see two images after this post; box =100 and box =150. Increasing the Money Mgt still works. The question of whether this is truly tbt testing, will hopefully be answered here or by IG/PRT. I’ll update as soon as I hear.

  I will get the VRT tests sorted too, once I’ve got the data to load into the “very user friendly” Excel correctly. I get gain and win graphs, but not scores?

  I also have a ML2 version from last week that so far looks good and I will post that. It uses ValueX for the boxsize and ValueY for the trailing stop loss, but its not graphing X and Y correctly – despite good profits and the fact that the X and Y algo codes are identical (apart from settings) to the correctly graphing Ehler’s Universal Oscillator posted earlier.

  The bottom line is that this system still makes profits (with a 100 and 150 boxsize), with certain date ranges, particularly in the last years and last few months, but until coffee returns to normal spreads, it’s probably better to use this Renko ML1 system on instruments like low spread currencies and indices. See image #3 below this post.

   

  

  Bard

  Participant

  Topics: 138

  Replies: 814

  Been thanked: 99 times
  04/23/2020 at 2:10 PM #127610
  Bard

  Relevant to my last comment.

  

  Bard

  Participant

  Topics: 138

  Replies: 814

  Been thanked: 99 times
  04/23/2020 at 2:32 PM #127617
  Vonasi

  Bard wrote:

  Wondered if you might know why that is @vonasi?

  I’ve not been following this topic very closely so am not really qualified to comment on the code. If you have the robustness code included in any strategy then the simplest way to test without it is simply  to add:

  tradeon = 1

  1	tradeon = 1

  at the end of the robustness section of code and just before the strategy part of the code. This way tradeon is always positive and the robustness test is nullified.

  Alternatively add something like this to your code and then you can simply turn the robustness tester on and off via one variable at the top of the code without the need to search through the code to find the right line to add or remove // from.

  RobustnessTest = 0 (robustness test code here) if not RobustnessTest then tradeon = 1 endif (strategy code here)

  1 2 3 4 5 6 7 8 9

  RobustnessTest = 0   (robustness test code here)   if not RobustnessTest then tradeon = 1 endif   (strategy code here)

   

  1 user thanked author for this post.

  GraHal

  

  Vonasi

  Moderator

  Topics: 192

  Replies: 5088

  Been thanked: 1811 times
  04/23/2020 at 2:49 PM #127618
  MAKSIDE

  Hi,

  thx for sharing this great system

  just for test, i tried to apply the valueX on trailingprofit on my 1S candle strategy. I hope to have good result after x weeks without this system.Then i will look it in order to optimize the parameters.

  system8 without and the other with machine learning

  What is your current feedback about it in demo ? with/without ?

  

  MAKSIDE

  Participant

  Topics: 37

  Replies: 343

  Been thanked: 98 times
  04/23/2020 at 3:12 PM #127625
  Bard

  Thanks Vonasi, yesterday I thought I had done your first suggestion and tried the tradeon = 1 and still got an optimisation window as per this screenshot below?
  Only way I could get it to stop was to Rem it out, so thanks for the extra code above.
  The question though, is why is the VRT code giving different results compared to a non VRT system with the same code? First two images here from my previous posts above: https://www.prorealcode.com/topic/machine-learning-in-proorder/page/15/#post-127605

  Cheers,

  

  Bard

  Participant

  Topics: 138

  Replies: 814

  Been thanked: 99 times
• Author
  Posts