Problem with ProBackTest, records disappear during optimization

• 10/18/2020 at 10:46 AM #147654
  IcaroFM

  Hi, I noticed that records disappear during optimization. I try to explain better: optimization begins, the first records sorted by gain show me for example 10k, 11k, 12k, after a while the record of 12k disappears, then 11k and at the end of optimization I find that the best result is 10k.

  It’s very strange. Has anyone noticed this behavior?

   

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/18/2020 at 10:54 AM #147655
  IcaroFM

  Is still running optimization.

  Here 2 screenshot that show the problem.

  Before and after.

  As you can see also 72117.90 record missing in “after” capture.

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/18/2020 at 11:07 AM #147659
  GraHal

  IcaroFM wrote:

  the record of 12k disappears, then 11k and at the end of optimization I find that the best result is 10k.

  We are limited to seeing only 100 results, but these should be the top 100 and so I cant see how 12K and 11K should dissappear and leave 10K … unless you are sorting by % wins or number of trades (instead of Gain)?

  

  GraHal

  Participant

  Topics: 214

  Replies: 8852

  Been thanked: 2215 times
  10/18/2020 at 11:12 AM #147660
  IcaroFM

  Hi GraHal,

  please take a look my second post.

  Optimization is still running and added another best gain, worst than the first one. I’m waiting for the final result.

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/18/2020 at 11:22 AM #147661
  GraHal

  IcaroFM wrote:

  As you can see also 72117.90 record missing in “after” capture.

  Only thing I can think of is the ‘display algorithm’ cuts out results if there is another result within x% … that way we get to see more overall variation of results within the limit of 100 results?

  Ironic if above is the case as it means PRT thought of user interests whereas in so many other functions the user seems to have been forgotten?

  I guess though it depends on who developed any particular function how ‘user centred’ that function is?  Shouldn’t be that way though.

  I’d best finish there as I’m moaning and going off topic! 🙁

   

  

  GraHal

  Participant

  Topics: 214

  Replies: 8852

  Been thanked: 2215 times
  10/18/2020 at 11:35 AM #147663
  IcaroFM

  GraHal wrote:

  IcaroFM wrote:

  As you can see also 72117.90 record missing in “after” capture.

  Only thing I can think of is the ‘display algorithm’ cuts out results if there is another result within x% … that way we get to see more overall variation of results within the limit of 100 results?

  Ironic if above is the case as it means PRT thought of user interests whereas in so many other functions the user seems to have been forgotten?

  I guess though it depends on who developed any particular function how ‘user centred’ that function is? Shouldn’t be that way though.

  I’d best finish there as I’m moaning and going off topic! 🙁

  In so many years of backtest it happened today and last Friday. I think there’s some problem in displaying the results. Let’s hope a moderator can help me. In any case, the last time I happened to save the screenshot at the end of the optimization I tried to enter the variable values of a missing record and the result is identical in terms of gain. I think there’s an hour left until the end of the back test (I have 2000000 bars): I’ll eventually post the final result.

  1 user thanked author for this post.

  GraHal

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/19/2020 at 8:00 AM #147697
  Nicolas

  Optimization made with PRT 10.3?

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20863

  Been thanked: 4775 times
  10/19/2020 at 8:13 AM #147704
  IcaroFM

  Nicolas wrote:

  Optimization made with PRT 10.3?

  Hi Nicolas, yes.

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/19/2020 at 8:19 AM #147707
  Nicolas

  Can you share a simple algorithm in order to replicate on my side please?

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20863

  Been thanked: 4775 times
  10/19/2020 at 8:40 AM #147708
  IcaroFM

  Here it is. Please try with Dow jones, Timeframe 2 minutes, 200k bars.

  Let me know, thank you

   

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/19/2020 at 10:00 AM #147721
  GraHal

  GraHal wrote:

  Only thing I can think of is the ‘display algorithm’ cuts out results if there is another result within x%

  Above suggestion is not the case as I can recall seeing loads of results (in my backtests) listed one under the other and all showing the same value for Gain (due to differing variable values).

  

  GraHal

  Participant

  Topics: 214

  Replies: 8852

  Been thanked: 2215 times
  10/19/2020 at 10:15 AM #147722
  GraHal

  I concur with  @IcaroFM findings, check …

  1.   Image 1 during backtest … shows lots of results at 1808.8 and 1797.7 … as highest Gain result.
  2.   Image 2 at end of backtest … shows only 1389.3 … as highest Gain result.

  What is going on??

  1 user thanked author for this post.

  IcaroFM

  

  GraHal

  Participant

  Topics: 214

  Replies: 8852

  Been thanked: 2215 times
  10/19/2020 at 10:34 AM #147727
  IcaroFM

  It’s the first time that it happened to me. But it’s the first time that I use an algo with multiframe function. I don’t know if the problem is linked to this new function.

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
  10/19/2020 at 11:03 AM #147734
  GraHal

  It could be happening to others, but most folks will be setting their optimiser going and then do something else (not watch results coming in).

  Let’s see if others read this and they check also and report back on here?

  

  GraHal

  Participant

  Topics: 214

  Replies: 8852

  Been thanked: 2215 times
  10/19/2020 at 11:30 AM #147735
  IcaroFM

  GraHal wrote:

  It could be happening to others, but most folks will be setting their optimiser going and then do something else (not watch results coming in).

   

  Yes, I think so. Maybe this is a historical bug and no one has ever noticed. i would like a prize if it were so 😀

  We wait for Nicolas’s checks

  

  IcaroFM

  Participant

  Topics: 12

  Replies: 41

  Been thanked: 2 times
• Author
  Posts