All (not yet) about Strategy Tester, Optimization and Cloud - page 17

[Ivan Titov]

Andrey Dik #:
then it is not clear why this happens, there are enough parameter options to distribute to agents

After I excluded network agents of the local computer where the test is running, the glitch seems to have disappeared. Maybe it's a coincidence, of course. But in any case you should not use them - there is really no performance gain.

[Ivan Titov]

I noticed one more glitch: after a genetic optimisation run, I start a single run by double-clicking on the best result and get a different result! How is this possible?

[Sergey Golubev]

How to Implement Auto Optimization in MQL5 Expert Advisors

Get ready to be introduced into the wonderful world of auto-optimizing forex trading algorithms. It can allow your Expert Advisor (EA) to adjust itself for the next iteration of trading based on how the market conditions are after a trade is done.

[Sergey Golubev]

The request -

Forum on trading, automated trading systems and testing of trading strategies

MQL5 Cloud Network: A Technological Breakthrough in Testing Trading Strategies

Robertomcat , 2024.10.31 17:12

What I'm saying is that this figure should be balanced by the processor's power. If I have a 10900K processor with 267 PR, and now I've installed a much more modern processor and it's limited to 300 PR, then that's unfair. The processor has a lot of power, and you have to pay for that power according to its PR.

Reply of MQ -

Forum on trading, automated trading systems and testing of trading strategies

MQL5 Cloud Network: A Technological Breakthrough in Testing Trading Strategies

Renat Fatkhullin , 2024.10.31 18:00

Yes, according to statistics it is clear that 14900 has reached 300 PR en masse.

We will raise the limit and do the analysis tomorrow.

Thanks for the information!

--------------------------

And finally -

Forum on trading, automated trading systems and testing of trading strategies

MQL5 Cloud Network: A Technological Breakthrough in Testing Trading Strategies

Renat Fatkhullin , 2024.11.01 13:10

Today we updated MQL5 Cloud Network to version 4651.

All testing agents will be updated automatically and after some time (up to 24 hours) their ratings will be recalculated.

--------------------------

More to follow: the question -

Forum on trading, automated trading systems and testing of trading strategies

MQL5 Cloud Network: A Technological Breakthrough in Testing Trading Strategies

Robertomcat , 2024.11.01 13:17

Okay, Renat, thank you very much.

According to the statistics you mentioned in the previous message, which processors currently provide the highest performance and PR?

The reply:

Forum on trading, automated trading systems and testing of trading strategies

MQL5 Cloud Network: A Technological Breakthrough in Testing Trading Strategies

Renat Fatkhullin , 2024.11.01 17:02

To the maximum:

• Intel Core i9-14900K
  
• Intel Core i9-13900K
  
• AMD Ryzen 9 7950X3D 16-Core
  
• AMD Ryzen 9 7950X 16-Core

It is important that the memory speed makes a significant contribution. Currently, Intel supports faster memory and this is (may be) the reason for the slightly better benchmark.

--------------------------

More - in the end of this page: https://cloud.mql5.com/en/stats

[Sergey Golubev]

More information -

Forum on trading, automated trading systems and testing of trading strategies

MQL5 Cloud Network: A Technological Breakthrough in Testing Trading Strategies

Renat Fatkhullin , 2024.11.03 17:46

New processors are coming soon.

We will upgrade the PR calculation to better differentiate processors and configurations.

In the computational network, the number of tasks received is affected by the available memory per core. You need at least 2 GB per core, and preferably more.

Therefore, if there is a lack of RAM, it is better to reduce the number of agents.

[Sergey Golubev]

Forum on trading, automated trading systems and testing of trading strategies

EA looks into the future on the tester. How to recognize?

MrBrooklin , 2024.11.12 05:37

Good question. In those distant times, when I tried to use other people's EAs, I was tormented by exactly the same question. The only test that gave a more or less accurate answer was Forward in the MT5 terminal . You do optimization, the period on history, for example, last year, and for Forward, set the period - the current year.

MANDATORY CONDITION!!! To conduct tests, you need to set the modeling mode " Every tick based on real ticks" .

There is another feature that quickly breaks the beautiful pictures after getting good results on optimization. You need to run the test several times with the same settings, but set the beginning of the testing period, for example, not from the 1st of some month, but choose days RANDOMLY. Five or six tests with randomly set dates and the picture will be clear. Good luck!

Best regards, Vladimir.

[Sergey Golubev]

A New Approach to Custom Criteria in Optimizations (Part 1): Examples of Activation Functions

The quest for the ideal optimization to find the right combination of parameters continues. The forum is awash with suggested methods to persuade the MetaTrader Optimizer to return and rank the various passes to enable a developer to choose a combination (or combinations) of parameters that are stable. The introduction of prop firms into the mix with their rigorous boundaries, understandably much more rigorous than might be required by a private trader, has made this even more testing.

[Sergey Golubev]

Manual Backtesting Made Easy: Building a Custom Toolkit for Strategy Tester in MQL5

Backtesting trading strategies is a cornerstone of successful trading, but automating every idea can feel restrictive, while manual testing often lacks structure and precision. What if you could combine the control of manual trading with the power of MetaTrader 5’s Strategy Tester? In this article, we introduce a custom MetaQuotes Language 5 (MQL5) Expert Advisor (EA) that transforms manual backtesting into an intuitive, efficient process—equipping you with a toolkit to test strategies on your terms.

[Sergey Golubev]

Forum on trading, automated trading systems and testing of trading strategies

New MetaTrader 5 platform version build 5120: improvements and fixes

Renat Fatkhullin , 2025.06.21 09:44

I will answer in the form of a summary:

1. The terminals detect all cores and allow them to be used in the tester:
   
   
   
2. The tester, including the cloud, also detects all cores and allows them to be used
   
   
   
3. Ideally, on pure CPU bound tasks, hyperthreading gives ~25% gain.
   
   The task of testing robots is not a purely CPU-dependent process. Other resources have a very strong influence: disk, memory, system synchronizations of the operating system.
   
   That is, using 72 logical cores instead of 36 physical ones, you create a double load with normal robots, but in the limit with the freedom/speed of disk and memory resources, you can ideally get 25% acceleration. In reality, you can get the opposite, a slowdown.
   
   There is no need to give an example of a "mathematical" mode without disks, data history and memory. This is beyond the scope of the issue under discussion and is clear.
   
   
4. RAM is the most critical resource for multi-threaded testing.
   
   If it's really simple, without discussing the real needs of a specific robot, then you need to have 2 GB per working core. Using a swap file for the sake of maximizing core usage is pure suicide. The lags will be incredible by definition.
   
   
5. In MQL5 Cloud Network we achieved very good and stable results when we updated the resource estimation of available agents
   
   Unfortunately, there were quite a few participants who registered "on all cores" with poor available memory, disk speed and network speed.

Recommendations:

1. Experiment in your cases on:
   - all cores
   - only on physical
   - on the number of cores so that there is enough memory
   
   
2. Do not use all cores, as their full loading causes general degradation of the operating system and all processes.
   
   
3. Don't try to boil everything down to one answer or requirement.
   
   You run unknown programs with unknown behavior and without any guarantee of efficient code. These programs can do any inefficient actions, which results in underutilization of the CPU or, on the contrary, useless loading of the processor, disk or memory.
   
   Therefore, you can't put everything down to the tester. Publish comprehensive technical evidence if you think some tester behavior is wrong.
   
   
4. When discussing technically complex processes, it is useless to use everyday logic

[Sergey Golubev]

Forum on trading, automated trading systems and testing trading strategies

Does MT5 support hybrid e-core yet ?

Eleni Anna Branou, 2025.08.03 17:26

After the latest 5200 MT5 build update, in order to take advantage of all your processor's cores and threads in the Strategy Tester (locally), go to MT5 >> View >> Strategy Tester >> Agents >> Local Network Farm >> right click >> + Add >> Add Agents from Host (automatically) >> Next >> Done.