MetaTrader 5 Strategy Tester and MQL5 Cloud Network - page 19

New comment
 
Karlson:

I am wondering if the genetic algorithm for the inputs will work if I use the Custom mode for optimisation and the code is as follows:

The algorithm doesn't seem to care. Although the question is interesting. Do you imply non-linear behavior of the optimization curve, or rather "out of fire and into the fire"? Try to check it yourself, I think it depends on the trading system.
 
Most likely to be knocked down by such a rough cut-off, painstakingly nurtured genes can often be thrown out by such conditions.
 

Yes, the question is interesting, I thank you for your opinion, I will have to check. For example, I already have indicators satisfying the system through complete optimizations, "manual" analysis, etc. Let me see whether the same conclusions will be reached by the tester or so.

Three parameters A, B , C. The best I recognize B = 12.In the process of optimization, my B = 12 will be in conjunction with other suboptimal A and C. The result will be reset output RF = -1.Thereby, after a while, my B = 12 will generally drop out of the field of acceptable values ...

Let's search... :-)

 
Renat:
Most likely to be knocked down by such a rough cut-off, painstakingly nurtured genes can often be thrown out by such conditions.
As my practice shows (on several EAs), nothing bad happens. I also cut off passes with less than 30 trades and degree in the target f-fi. And very seldom the most successful parameters are concentrated in the range of 30 trades. Well, what's the use of meticulously cultivated genes if they do not lead to the desired result. All of this is IMHO, I haven't read the maths on this subject, it's a subjective feeling.
 

+ wish: please make it possible to win back your agents from Cloud (for example, if agent received connection with correct username/password, it does not take new tasks from Cloud for X minutes, but finishes current task and waits for connection; if X minutes passed - it requests tasks from Cloud again) - now it is almost impossible to win back your agents - you have to stop and start them manually, and reconnect them in tester.

+ wish: agents inactivity time before they take tasks from Cloud is clearly less than 5 minutes (as stated in the beginning) - please return 5 minutes or point out the reason why it was removed.

 
notused:

+ wish: please make it possible to win back your agents from Cloud (for example, if agent received connection with correct username/password, it does not take new tasks from Cloud for X minutes, but finishes current task and waits for connection; if X minutes passed - it requests tasks from Cloud again) - now it is almost impossible to win back your agents - you have to stop and start them manually, and reconnect them in tester.

Yes, this will definitely be done. There will be a timeout of 5 minutes to connect to the claude after the local task is taken.


+ wish: agents idle time before they take tasks from Cloud - clearly less than 5 minutes (as stated in the beginning) - please return 5 minutes or poke your nose in where the reason why it was removed.

Not removed, but we will check.
 

Reread two threads on the subject, more than 30 pages in total, totally enthralled...

Got my agents on board.

Two questions remain.

Has the service been launched with payment?

Will the money be able to spend it only on my testing or will it be real money with possibility of withdrawal or payment for "Jobs" service?

 
IvanIvanov:

Reread two threads on the subject, more than 30 pages in total, totally enthralled...

Got my agents on board.

Two questions remain.

Has the service been launched with payment?

Will the money be able to be spent only on my testing or will it be real money with possibility of withdrawal or payment for "Jobs" service?

1. With payment not yet, it seems in September will begin full work (for the period of testing developers charge a bonus).

2. Possible: withdrawal + payment for goods in the shop + payment for work + payment for cloud agents,

 
Karlson:

Greetings!!! I chose Balance+Recovery Factor as the main optimization criterion. Sometimes, with wide input data the number of trades decreases, and the Factor is high enough. For my system this is the wrong way. The number should remain in a certain range, and already in this area to look for combinations. I am interested, won't the genetic algorithm of input data selection, if I use Custom mode for optimization, and the code will be as follows :

It is better to make not cut-off, but punish by deduction of certain number (what number to deduct set by experience).

In this way the bad results will retain information about evolution of the system. Thereby you simply decrease the area of bad results, but keep the relief.

In your case, if Genetics gets into the bad results area, it is difficult to get out of there, as there is no information where to move, all the bad results equal -1. This is a plane that has no preferred points.

 

My agent activity dropped to zero after the site update - is this normal?

And another question - does the amount of history on the computer affect agent activity?

And does MT have to be logged in at least on the demo account for agents to work?

1...121314151617181920212223242526...42
New comment