ForumSections

to mt5 support , bug news releases mt5 (1940+) - page 3

New comment
James Cater:

...

I believe this build 1949 has been pushed out to the brokers

I don't think so, I have 3 different brokers where it's still 1940.

Alain Verleyen:
Hopefully 1948 which is now available, is working too ;-)
Another error "Access Violation" in build 1950. It's getting rare, but still happening.
Gustavo Hennemann:
Another error "Access Violation" in build 1950. It's getting rare, but still happening.
Are you able to isolate when it happens to eventually reproduce it ?

Mr.Alain,


Since april/2018  with new mt5 releases and windows 10 64 bits with falls creator update the ontick  freeze my machine.. same problems with anothers friends in brazil that are using windows 10 64 bits..

but I'm want to relate a new bug while optimizing , I called it "the OnInit bug with the last "build 1950"


I'm optimizing using OHLC (well this the only "mode" that works without hang my windows 10 with amd octare processor)

ok,

how reproduce it.
=================
my EA has a routine in the OnInit to "exclude" bad values using return(INIT_FAILED) on the "OnInit" of the EA
to economize time in the optimization, this works with the old version but now it's breaking the optimization.

I started the optimization OHLC, 
and I go to the "diary" Tabs and see the messages , that are normal... because I'm excluding the bad values from the OnInit



but while it is in  the pass 512 (or a little more) the mt5 stopped the optimization without conclude it (it doesn't complet at the end) 

and the agents are hanged with 0% cpu usage, the optimization is "forced to end" and the "start" but still GRAYED...

I have to go to the windows taskmanager

and I kill manually the 8 agents hanged, and after it the "START" button in the in the mt5 comes "ENABLED" again...


Files:
bug.png  93 kb
bug1.png  64 kb
zemo:

I am afraid you are excluding too much parameters and the genetic algorithm is lost.

Anyway, it's not very useful to report problems without providing all needed information to reproduce it. I don't have MT5 backtests slow or crashing/hanging, so I can't help without a way to reproduce the problem.

see how it does,

the correctly was to "kill the agents as the normal way in the old releases" ,  enable "START" button...

but this 1950 release after "stop" the genetic test it "leave" mt5 agents ON... 

Files:
bug3.png.jpg  161 kb
bug4.png  26 kb
bug5.png  73 kb
Alain Verleyen:
Are you able to isolate when it happens to eventually reproduce it ?
I tried everything but the problem looks like random.
Alain Verleyen:

I confirm. On my side seems always called but with huge delay (more than 3 minutes !).

Fixed in build 1953.
Gustavo Hennemann:
I tried everything but the problem looks like random.
I got one after using delete on an invalid pointer, don't remember the build.
zemo :
...

You are it's an MT5 bug, as the Start button (among others) remains grayed and unusable. I confirm it and it's still present in build 1953.

However the way you are excluding parameters with the Genetic Algorithm is not good, you will never be able to use GA this way with MT5. You need to change your approach to exclude less than 25% of the combinations.

Forum on trading, automated trading systems and testing trading strategies

Today the MT5 terminal has been updated and the Optimization window is not shown during the test.

Renat Fatkhullin , 2018.07.02 23:18

I recommend for genetics:

  1. Read about the mechanics of genetic search
  2. Choose a sufficiently smooth objective function to enable the algorithm to see improvements over generations and move forward. Do not be fooled by false results. The genetic algorithm is quite a thin and sensitive thing.
  3. Do not kill through INIT_PARAMETERS_INCORRECT more than a quarter of the population per pass (use in critical cases, but it is better not to use at all). A quarter is a rough estimate and is by no means a recommendation or permission to do so.
  4. The enumerated parameters must be operational throughout the span. Do not make known holes in the iterated parameters, so as not to generate INIT_PARAMETERS_INCORRECT. You’re doing it solely for your convenience, and you’re not thinking about the brute force process.
Yes, you need to rebuild the logic of parameters under the process of searching and the search method used, and not under your convenience.
Reported on Russian forum.
1234567
New comment