Question for developers - using all computational cores during optimisation - page 5
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
The problem with distributing tasks evenly is also related to the order in which agents are activated. This is especially obvious when adding new agents, after running optimisation.
My hope is that EAs and their companion files will, in the bright future, be uploaded once to a remote computer with agents, which will save time, and make the distribution of computing resources more even.
Question to MT developers: When will there be any progress? Please tell me when the catastrophic errors described above will be fixed, because of which optimisation does not actually work ....
Question to MT developers: When will there be any progress? Could you please tell us the deadline for fixing the catastrophic bugs described above, due to which the optimization actually doesn't work ....?
What "catastrophic mistakes" are you referring to?
How come the optimization does not work?
>What "catastrophic errors" are you referring to?
>And how this optimization doesn't work
Everything is described above and not only by me.
1. Network agents don't actually work, they just don't start
2. When calculating a new set, it takes a very long time and thinks about something.
3. It distributes packets of tasks only to one and the same cores when there are many other free cores
Result: optimizations take up to 4 times longer to run than they used to ....
Still no answers to the questions above
but everything seems to be working fine after the updates, although there are still some idle cores
the issue of inadequate distribution of jobs to cores remains:
1. why different cores of network agent are given different number of jobs and not each core receives the same job
2. What will happen if I disable agent which has 58 jobs at a given time like on the picture above, will they still be calculated or will they be forgotten and never calculated?
Hello, trying to speed up testing using TestorStop() function in Opticket by stopping individual testing if EA reaches unacceptable values. As a result, some runs are faster than others and one of the agents finishes first (somewhere i read that if a testing agent is idle for some time, it gets turned off) - when all other agents finish, tasks are not sent to the first agent. And so test agents drop out one by one, until only one agent remains - in this case we can't talk about speed. I.e. the speed can be increased, but the technical errors of implementation of distributed computing do not allow for it. If we runTestorStop() on OnInit(), for example, to optimize 2 parameters, one of which should be greater than the other, during the first run the speed increases too, but after that test agents stop doing tasks, although the counter of sent tasks increase with a great speed, while the counter of solved tasks remains idle. Maybe someone knows how to bypass this - the topic is interesting, the speed of testing by discarding unnecessary results in 10 times greater than normal testing, in addition, zeroing unnecessary results in OnTest() can steer the genetic algorithm in the right direction for us!
Developers, solve the problem of handing out tasks .... the "one core - one job" principle
I have 10 cores of main CPU counting now since they got 40+ jobs and another 40 network cores idle since they counted them all ... you can distribute the jobs evenly.... it's elementary simple for programming
Developers, solve the problem of handing out tasks .... the "one core - one job" principle
I have 10 cores of main CPU counting as they get 40+ jobs and another 40 network cores idle as they counted them all ... jobs can be evenly distributed .... it's elementary simple for programming
This is not the right approach - you don't have to give out jobs one by one, you have to redistribute the capacity if there are free resources, i.e. cancel the jobs already given out and give them to others to be executed. At the same time, it is necessary to analyze the performance of each agent in order to give the kernel the right number of new jobs for execution.