MT5 terminal updated today and the "Optimisation" window does not show up during the test - page 3

New comment
 
Aliaksandr Yemialyanau:

The optimisation schedule is not updated.

2500 passes:

In genetics or regular passes?

Does it not update or is there a difference in the counters of the results shown and the number of passes declared?


The graph is updated, only already in large packs, not in rltime. Priority is given to useful work, not to rltime visualisation.

We don't output knowingly bad results on graphs, because they would just spoil the whole visualization. After all, most of the optimization options will push the balance to zero. Therefore, you will see on the graph a counter of more or less decent results.


And let me remind you that we do not accept pictures without environment description. Please be more responsible with your questions.
 

What else is new is the new mechanism for handling the cache of previous results in genetics.

We now show all previous genetics passes when displaying results. This gives a better understanding of the search process in genetic optimization.

Below is a graph on the fifth run of genetics:

We can see that on the fourth run the randomizer has broken out of the local extremum and found a new slide of better options.

When using genetics it is important to understand the mechanics of genetic search and remember to run the genetics multiple times over the same parameters so that the randomizer can more fully cover the search area.

With the new tester, this becomes easier to understand and see by eye.

 
Renat Fatkhullin:

In genetics or ordinary passages?

It doesn't update or there is a difference between the counts of the displayed results and the reported number of passes?


The graph is being updated, only already in big batches, not in rltime. Priority is given to useful work, not to rltime visualisation.

We don't output knowingly bad results on graphs, because they would just spoil the whole visualization. After all, most of the optimization options will push the balance to zero. Therefore, you will see on the graph a counter of more or less decent results.


And just a reminder - we do not accept pictures without a description of the environment to be reproduced. Please answer your questions responsibly.

Genetics. The same graph after 1000, 2000, 3000 passes.

After optimization is stopped, the graph is completely different:

"Custom max" should be optimized, but judging by the graph and the results, it's either balance or something else that's being optimized. The graph does not match the optimization results.

 
Renat Fatkhullin:

We can see that on the fourth run the randomiser broke out of the local extremum and found a new slide of better options.

When using genetics it is important to understand the mechanics of the genetic search and remember to run the genetics multiple times on the same parameters so that the randomizer can more fully cover the search area.

With the new tester it becomes easier to understand and see with your eyes.

Yes, interesting feature, useful !

Bravo. Looking forward to an update of the terminal.

 
Aliaksandr Yemialyanau:

Genetics. The same graph after 1000, 2000, 3000 passes.

After optimization is stopped, the graph is completely different:

"Custom max" should be optimised, but judging by the graph and the results, it's either balance or something else that's optimised. The graph doesn't correspond to the optimization results.

I don't see any problem with the tester, but I do see a problem:

  • improper use of the Expert Advisor
  • wrong conditions
  • the Expert Advisor is simply unprofitable, which makes it impossible for the tester to find anything at all
  • wrong code of custom condition.
    yes, this is the obvious problem - you can't use any such primitive parameter as an optimization target point. you are killing the genetic algorithm by cheating it!
    double OnTester()
  {
//---
   return(TesterStatistics(STAT_RECOVERY_FACTOR));
//---
  }
    If you don't want to use profit as an optimization function, you need to invent your own complex multifactor formula with weights, sigmoid cutoffs etc.

  • misunderstanding what custom max(recovery factor) means on the graph and why almost all results are at the bottom.
  • lack of comparison with the previous version of the terminal, which would remove some of the questions


In general, the tester has nothing to do with your question. It is about your misunderstanding of the genetic optimization method and how it should be properly used.

After all, it is enough to set optimization by "Balance Max" and the tester suddenly starts to draw and optimize everything correctly, isn't it?


 
Georgiy Merts:

Yes, interesting feature, useful !

Bravo. Looking forward to terminal update.

Update has been available on MetaQuotes-Demo for a long time.
 
Renat Fatkhullin:

I don't see any problem with the tester, but I do see a problem:

  • improper use of the examiner
  • wrong conditions
  • The Expert Advisor is simply loss-making, which makes it impossible for the tester to find anything at all.
  • wrong code of custom condition.
    yes, this is the obvious problem - you can't use any such primitive parameter as an optimization target point. you are killing the genetic algorithm by cheating it!
    If you don't want to use profit as an optimization function, you need to invent your own complex multifactor formula with weights, sigmoid cutoffs etc.

  • misunderstanding what custom max(recovery factor) means on the graph and why almost all results are at the bottom.
  • absence of comparison to the previous terminal version, which would remove some questions


In general, the tester has nothing to do with your question. It is about your misunderstanding of the genetic optimization method and how it should be properly used.

After all, you just need to set the optimization by "Balance Max" and the tester suddenly starts drawing and optimizing everything correctly, doesn't it?


This Expert Advisor optimized perfectly in previous versions of the terminal.


After restarting the optimization, the chart started showing correct values:

And then again wrong values:


The results (optimization tab) do not show such large values as on the chart.

 

Yes, I ran your Expert Advisor in genetics and got values on the chart that are not in the results table. Probably, the automatic mode of hiding results that do not pay profit has triggered.

Plus there was a problem with periodic updates to the optimisation graph.

We will investigate everything on Monday and I will report on the results.

 
MT5 periodically closes, windows_7 writes that there is an error in these files:
Files:
Er.zip  351 kb
 
Renat Fatkhullin:

No, we won't.

I explained above why pulling 500,000, 1,000,000, 10,000,000, 50,000,000 into a table of useless intermediate rows for the human eye makes no sense. It is very expensive technically and completely useless.

@Renat Fatkhullin The intermediate rows in genetics are not millions and spend fewer resources on sorting, and the human eye is quite able to digest the table, leave it for genetics at least.

The optimization graph window is always available, it is more economical for resources (it doesn't require re-sorting) and allows you to observe the optimization process in a human-eye friendly way.

Again, leave the old on-line rendering of the optimization graph for genetics rather than a batch by total, not that many of those resources are consumed.


Forgot another important point in what's done:

  • Dramatically reduced memory consumption by the terminal during heavy optimizations
12345678910...25
New comment