OpenCl and the tools for it. Reviews and impressions. - page 15

 
Reshetov:
I see. I believe. Vanya, I mean, Renat, I want one too.

You have lost the argument.

You are out of touch with the iron, out of touch with the cludes, out of touch with the business of developing trading platforms, which gives a legitimate result.

 
Renat:

Here we go again with the cheating.

You have to prove through screenshots if words don't work.

Don't you know the peculiarities of genetics and the concept of population size? Enabling genetics mode and generating tasks from 64 to 256 are great conditions for finding thousands of agents.

Take a look at the just discussed topic Discussion of the article "MQL5 Cloud Network speeds up calculations" - the testing mode of genetics in the cloud has just been analyzed there.


So with the genetic algorithm, Clouds Network won't go to full performance?

Thanks for the article, I'll read it for sure.

Unfortunately for me, I don't need to go through all the options. Although the reason why the network is not working at full capacity has been worked out.

 
Reshetov:

So with the genetic algorithm Clouds Network will not reach full performance?

Not precisely because of its genetic brute force ideology - to generate short populations (64-256) for the next crossing.

On the example given by you it is clear that claud network even in this case gave speed-up at least 7 times by number of passes 3 455 / 507 = 6.8 times, and much more in time (computers in claud are notoriously more powerful).


Thanks for the article, I will read it for sure.

Kindly correct the factual errors in your earlier post about claud network.
 
Renat:
Kindly correct the factual errors in your earlier post about the Cloud Network.

I'll start by reading the article carefully.

Because if it turns out that Cloud Network really can't give performance with a genetic algorithm, I'll add this note to my posts.

 
Renat:

Those who agreed with me in this thread also haven't read the article and of course they got the same result when I turned on genetics.

I.e. there should be a clue in red letters in the terminal that MT5 with genetics enabled is inferior in performance even with distributed calculations to MT4, and with genetics disabled one should not forget that distributed calculations are not free. In other words, it is no longer possible to experiment with trading systems like on MT4 - you have to pay for everything.

 
Reshetov:

The genetics are sorted out. Why do I see fewer agents online than you do?

Because I write clear and detailed explanations, but not everyone reads them.

The main task of a giant network is to sleep. To sleep whenever there is a chance, so as not to waste resources. That's why all network components explicitly save all resources (network, traffic, cpu, etc) and instantly shut down when there are no tasks.

Cloud servers send as many agents to sleep as possible, which results in 90-99% of the agent network being invisible. All agents are raised and assigned to tasks strictly on the appearance of the next task.

That is, 100-200 "available" agents that are visible at rest are only those who were in a "connected to ask about task availability" state at the time of the last poll. As soon as tasks appear, instantly the cloud server (which is not a simple dispatcher) starts allocating agents. And the allocation of agents goes in a kind of competitive mode, when tasks from different customers are distributed more or less rationally.


Also, I wonder why there are only four agents on the forward tests, since genetics is turned off there?

I guess you mean "4 cloud servers", not "4 agents"?

In full brute-force (regardless of forward analysis mode) all cludes are used, and in genetics only one cluded server is used (it doesn't make sense to disperse micro-packages into 64-256 tasks).

 

Renat:


Because I write clear and detailed explanations, but not everyone reads them.

There is such a letter.

Renat:


Full brute force (regardless of forward analysis mode) uses all cludes, but in genetics only one cludes server (it doesn't make sense to disperse micro packets in 64-256 tasks).

Pity. And on local agents MT5 (compared to MT4) doesn't even pull the weakest calculations, without powerful iron. The only misunderstanding comes out. Everything that is not required is there and working at full power, while the most demanded is in such a state that there is no sense even to specify in what state it is. It's not a platform, it's absurd.

I have corrected my posts so as not to cause misunderstandings.

 
Reshetov:

Those who agreed with me in this thread have also not read the article and of course, turning on the genetics got about the same result.

Don't take silence for consent.

If I hadn't stopped your stream of consciousness, you would have gone far away. Unfortunately, I don't always have enough time to educate people. So I have to periodically catch some of them one by one and flog them in public.

That is, I need to print somewhere in the terminal a hint in red letters that MT5 with genetics enabled is inferior to MT4 even with distributed calculations and with genetics disabled one should not forget that distributed calculations are not free. I.e. it is no longer possible to experiment with trading systems like on MT4 - you have to pay for everything.

Well right now we'll go and write comparisons to the delight of your punctiliousness. You have fought for it, you are obsessed with MetaTrader 4.

Don't be fooled - you are not personally in a position to read it:


Many people try not to accept changes, remaining within the framework of previous knowledge.

When changes come, they do not want to acknowledge them and defend their positions with all their might, even with blatantly weak arguments. In the case of MetaTrader 5, this is exactly what is happening. In the four years that we spent developing it, it is head and shoulders ahead of MetaTrader 4.

 

Renat:

In the case of MetaTrader 5, this is exactly what is happening. In the 4 years we've spent developing it, it's head and shoulders ahead of MetaTrader 4.

In terms of functionality, yes, MT5 is ahead of MT4.

As to devouring of resources, both own and other people's, MT5 has very few competitors. MT4 has a rest in this respect.

 

Renat, are there any other code examples using OpenCL in MQL5?

I want to code a neural network using OpenCL and post it as a test, as I did in this thread earlier. For me, the more examples the better, OpenCL & MQL5 examples.

Reason: