Remote control of strategy tester | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Remote control?? 3 replies

strategy builder/tester for new scalping strategy 1 reply

Free or Good remote desktop control? 15 replies

2009.03.11 15:41:15Tester memory handler: tester stopped because not enough memory 19 replies

mt4 strategy tester 0 replies

Remote control of strategy tester

  • #1
  • First Post: Edited 9:03pm Jan 29, 2011 9:01pm | Edited 9:03pm
Hello All,

I'd like to automatically pass inputs to and get results from the MT4 strategy tester and optimiser.

As an example, I'd like to:
1. optimise a strategy over a particular period
2. load the results in an external program and choose a set of parameters to use
3. run the backtester with the new optimised parameters and analyse the results
4. change the tester period (move it forward in time) and test the strategy
5. load the results back into the external program
etc etc.

Ideally I'd like to use the external program (c# or c++) to control the entire process so that the process can be repeated numerous times without any intervention.

Does anyone have any experience with remotely controlling the strategy tester, or do you know if it is even possible?

Thanks,

Michael
  • #2
  • Jan 30, 2011 5:36am Jan 30, 2011 5:36am
Quoting PipMasterMik
Hello All,

I'd like to automatically pass inputs to and get results from the MT4 strategy tester and optimiser.

As an example, I'd like to:
1....
Yes, it's possible, I am just working this way.
The external program I'm using is written in C#.
You can start the MT4 terminal and automatically run the backtester from the command line. Give a look at the documentation of your MT4 client.
  • #3
  • Feb 1, 2011 9:39pm Feb 1, 2011 9:39pm
Quoting asasa
Yes, it's possible, I am just working this way.
The external program I'm using is written in C#.
You can start the MT4 terminal and automatically run the backtester from the command line. Give a look at the documentation of your MT4 client.
Thanks for the response asasa. I looked and looked for this documentation and eventually found it within the MT4 help application itself; Thanks for the tip!

I noticed that the exported optimisation reports do not include the settings for the optimisation, only the pass number and the results. Do you somehow link the actual parameters with the optimisation output in your application? Or do you simply reconfigure for numerous backtests at the input parameters of your choosing?

It also looks like there is a lot of parsing involved when interpreting the HTML; is this how you do it or is their an easier way (i.e. export the file in a more readable CSV format)?

Thanks once again,

Michael.
  • #4
  • Feb 2, 2011 1:07am Feb 2, 2011 1:07am
Quoting PipMasterMik
Thanks for the response asasa....

In case you have missed it, I refer to the documentation you can find in your MT4 client, go to Help -> Help topics -> Tools -> Configuration at Startup
With this approach you don't need the MT4 optimization feature any more, all the runs are defined and managed by the external application.
My external application parses the html report to extract the backtest results.
You can also implement another approach, in the deinit() function your EA can write the backtest results to a CSV file.
In this case all the backtest stats must be manually calculated by the EA itself.
  • #5
  • Feb 2, 2011 4:24am Feb 2, 2011 4:24am
Quoting asasa
In case you have missed it, I refer to the documentation you can find in your MT4 client, go to Help -> Help topics -> Tools -> Configuration at Startup
Thanks, it took me a while to find it but that's where I eventually got it from!

Quote
With this approach you don't need the MT4 optimization feature any more, all the runs are defined and managed by the external application.
That makes perfect sense. It has the advantage that you can design your own optimisation algorithm, as the MT4 one can sometimes be limited.

Quote
You can also implement another approach, in the deinit() function your EA can write the backtest results to a CSV file.
In this case all the backtest stats must be manually calculated by the EA itself.
I hadn't thought of this, good idea and simple to implement. This would work well for me because I would like to use my own custom fitness and filtering function that the stats don't really help much with anyway.

Are you aware if there is much overhead in repeatedly restarting the MT4 application every time you feed it a single set of parameters? Does the restart take a few seconds each time?

If the deinit() function is called on every pass of the optimiser, I guess I could also write the optimisation parameters being tested along with the trade results to a new file on every pass of the optimiser. This would eliminate any overhead from having to repeatedly restart the application on every individual backtest.

Sorry for all the questions; I'm currently in the midst of deciding which of two platforms to use:

Option 1: Using MT4 to do all trade logic and data processing and use a c# application to analyse results and provide EA inputs (I expect that this is close to what you are doing?)

Option 2: Importing market data from an EA into my c# application, implementing all trade logic, backtesting and optimisation in my own custom engine, and sending back instructions to the EA to place trades etc.

Option 2 will be a lot more work, but it will probably be much more efficient in backtesting and hence could actually have a faster strategy development time (my trading system requires a lot of optimisation passes).

Cheers,

Michael
  • #6
  • Feb 2, 2011 7:27am Feb 2, 2011 7:27am
Quote
Are you aware if there is much overhead in repeatedly restarting the MT4 application
every time you feed it a single set of parameters?
Does the restart take a few seconds each time?
It takes the same time you usually have to wait when you start your terminal and then the backtest by hand.
At the end of the backtest the terminal automatically exits, leaving a html report with the results of the run.

Quote
If the deinit() function is called on every pass of the optimiser, I guess I could also write the optimisation
parameters being tested along with the trade results to a new file on every pass of the optimiser.
This would eliminate any overhead from having to repeatedly restart the application on every individual backtest.
This overhead of a few seconds usually is not that important, since the backtest run itself takes several minutes.
I suggest you first of all to optimize your EA for fast execution.
How long does a run of yours last?

Quote
Option 1: Using MT4 to do all trade logic and data processing and use a c# application to analyse results and
provide EA inputs (I expect that this is close to what you are doing?)
Option 2: Importing market data from an EA into my c# application, implementing all trade logic, backtesting
and optimisation in my own custom engine, and sending back instructions to the EA to place trades etc.
Option 1 or 2 depends on your needs and time requirements.
Option 2 is ok if your strategy is well defined and stable.
If you are still working on the strategy itself, then option 1 is preferable.

Another point to think on, you should provide good tick data to the backtester.
  • #7
  • Feb 3, 2011 5:00am Feb 3, 2011 5:00am
Quoting asasa
This overhead of a few seconds usually is not that important, since the backtest run itself takes several minutes.
I suggest you first of all to optimize your EA for fast execution.
How long does a run of yours last?
Thanks once again for the response. My strategies are very lean in terms of processing effort and I use relatively short optimisation periods, leading to around one optimisation pass per second (or maybe a few seconds max). However, I do a lot of optimisation passes, so the overhead of restarting the terminal for each optimisation pass will probably add significant overhead. A rough complete optimisation can be completed overnight.
  • #8
  • Feb 3, 2011 7:22am Feb 3, 2011 7:22am
Quoting PipMasterMik
Thanks once again for the response. My strategies are very lean in terms of processing effort and I use relatively short optimisation periods, leading to around one optimisation pass per second (or maybe a few seconds max). However, I do a lot of optimisation passes, so the overhead of restarting the terminal for each optimisation pass will probably add significant overhead. A rough complete optimisation can be completed overnight.
Then in your case the overhead will be relevant...
As far as I understand, you need to feed your EA with the results of the backtest, so why not implementing the optimization engine in the EA itself?
For example, you can write a C++ dll for MT4 for doing this at maximum speed (create a separate thread for the optimization procedure, so that your EA can continue to work in the meanwhile).
  • #9
  • Feb 3, 2011 5:42pm Feb 3, 2011 5:42pm
Quoting asasa
Then in your case the overhead will be relevant...
As far as I understand, you need to feed your EA with the results of the backtest, so why not implementing the optimization engine in the EA itself?
For example, you can write a C++ dll for MT4 for doing this at maximum speed (create a separate thread for the optimization procedure, so that your EA can continue to work in the meanwhile).
Yes, this is definitely another option. However, writing an optimisation engine will need to incorporate a backtester and trading logic, and if I've written that there is little advantage in having two sets of trading logic (i.e. one in the EA and one in the external application).

I think I'll write a generic EA that sends market data out and receives trading instructions back from an external application, and all other processing is done in the EA. I think you've helped make my pathway forward much clearer!

Out of interest, have you tried your suggested scheme of using a dll? I'm interested in what level of performance improvement there would be? My guess would be 100x or more.

Cheers,

Michael
  • #10
  • Feb 4, 2011 1:15am Feb 4, 2011 1:15am
Quoting PipMasterMik
Yes, this is definitely another option....
You are right, but in C++ the improvement in performance would be dramatic.
I think everything depends whether your ts is stable or still in development phase, and how much optimization speed is important for your trading style.
Right now I'm not using dynamic optimization because I don't need it.
  • #11
  • Feb 4, 2011 3:18am Feb 4, 2011 3:18am
Sorry, a mistake in my last post; I mean't that the processing will all be done in my external application (not the EA) to speed it all up.

I'm using dynamic optimisation but having said that it doesn't need to happen that often (daily would be sufficient). However, faster optimsation means I can actually backtest it in reasonable time, and also start running multiple pairs.

Are you interested in describing how your system works?
  • #12
  • Feb 5, 2011 6:02am Feb 5, 2011 6:02am
Quoting PipMasterMik
Sorry, a mistake in my last post; I mean't that the processing will all be done in my external application (not the EA) to speed it all up.

I'm using dynamic optimisation but having said that it doesn't need to happen that often (daily would be sufficient). However, faster optimsation means I can actually backtest it in reasonable time, and also start running multiple pairs.

Are you interested in describing how your system works?
I don't use dynamic optimization, I have written an external C# application to automatically download tick data (I always backtest with 99% quality), convert it in the correct format for MT4 and copy in the right locations.
It can also run, test and optimize a large set of parameters without the human intervention (this is done by invoking the MT4 terminal from the command line).
It collects the results, post-processes data and writes CSV files that can be directly opened in Excel.
The purpose is not to have the maximum profit, but solid results for a large variation of the input parameters.
I am planning to write my simulation engine in C for maximum backtest speed, but this is not my current priority.
  • #13
  • Feb 5, 2011 6:09am Feb 5, 2011 6:09am
Hey there asasa,

So you don't have to use the "Tools -> History Center -> Import" feature of MetaTrader? You just convert the data directly?

I'm also using external data for backtesting though it's M1, not tick data. So I'm getting by with 90% rather than 99%.
You are in a maze of twisty little passages, all alike.
  • #14
  • Feb 5, 2011 6:23am Feb 5, 2011 6:23am
Quoting stevegee58
Hey there asasa,

So you don't have to use the "Tools -> History Center -> Import" feature of MetaTrader? You just convert the data directly?

I'm also using external data for backtesting though it's M1, not tick data. So I'm getting by with 90% rather than 99%.
Yes, you are right.
I'm using Dukascopy tick data, my C# application automatically performs the steps described in Birt's site to download and convert it in MT4 format. It also copies the final data files in the correct directories and checks they are still there before running any backtest. At the end of the backtest it always verifies the simulation ran at 99% quality.
  • #15
  • Feb 6, 2011 5:19am Feb 6, 2011 5:19am
Quoting asasa
It can also run, test and optimize a large set of parameters without the human intervention (this is done by invoking the MT4 terminal from the command line).
It collects the results, post-processes data and writes CSV files that can be directly opened in Excel.
.
Thanks for the description. Do you then manually analyse the results in excel and then transfer these to your MT4 EA for live trading, or is this part also automated?
  • #16
  • Feb 6, 2011 1:11pm Feb 6, 2011 1:11pm
Quoting PipMasterMik
Thanks for the description. Do you then manually analyse the results in excel and then transfer these to your MT4 EA for live trading, or is this part also automated?
At the moment this step is done by hand, since I do not optimize very often.
  • #17
  • Aug 29, 2011 8:47pm Aug 29, 2011 8:47pm
I have a quick question...

How do you tell when the terminal.exe has exited after its run its test (strategy tester invoked via the command line)? Do you use a handle? I can't remember all of the .NET class libraries right off, and I'm not sure of the class used to launch another process.

But basically, if I used my own .NET app to launch terminal.exe, and have it run a backtest or optimization, how can I tell when terminal.exe has exited once the test run or optimization run has completed?

Thanks!
  • #18
  • Aug 30, 2011 1:54pm Aug 30, 2011 1:54pm
Quoting tonyc2a
I have a quick question...

How do you tell when the terminal.exe has exited after its run its test (strategy tester invoked via the command line)? Do you use a handle? I can't remember all of the .NET class libraries right off, and I'm not sure of the class used to launch another process.

But basically, if I used my own .NET app to launch terminal.exe, and have it run a backtest or optimization, how can I tell when terminal.exe has exited once the test run or optimization run has completed?

Thanks!
I periodically check "terminal.exe" amongst the active processes. Look for Process.GetProcesses() api.
  • #19
  • Aug 30, 2011 2:22pm Aug 30, 2011 2:22pm
Cool thanks! I can instantiate a new Process object, call the .Start() method and save the unique process id for checking (allows me to have multiple terminal.exe's running). And there is even an .Exited event. Perfect!

I was also thinking about using the deinit() function on the EA, and have it write a CSV file similar to what was described previously on this thread, and then invoking the FileSystemWatcher class to see when that file was created/modified. But the problem with this is if something happens to the terminal.exe process and the deint() function does not run.
  • #20
  • Aug 31, 2011 1:21am Aug 31, 2011 1:21am
Quoting tonyc2a
Cool thanks! I can instantiate a new Process object, call the .Start() method and save the unique process id for checking (allows me to have multiple terminal.exe's running). And there is even an .Exited event. Perfect!
I found that the .Exited event does not always work as expected.
I don't exactly remember the issues I had to face.
Feel free to experiment.
0 traders viewing now