Automated Trading Championship 2007: common mistakes in experts - page 3

New comment
 
The maximum allowable test run time is 5 minutes (on modern Xeon, AMD X2).
OK, esteemed jury, are there any results from any more or less available benchmark that matches the calculations for a test run on these modern computers? Knowing my own benchmark, I will be able to estimate how many times my run times can be higher to pass the test in terms of time.
 
Mathemat:
The maximum allowable test run time is 5 minutes (on modern Xeon, AMD X2).
OK, esteemed jury, are there any results from any more or less available benchmark that matches the calculations for a test run on these modern computers? Knowing my own benchmark, I will be able to estimate how many times my run times could be higher to pass the test in terms of time.
The tests are currently running on an AMD X2 4800, 4Gb RAM.
 
And how many times faster does this monster benchmark against my old PIV Northwood 2.4 GHz + RAM 512 MB? Actually, that's why I asked about the benchmark, so as not to get similar questions from others...
 

Hi all.

I also had the problem of 5 mins because of the indicators which were always calculating on every tick and the time was 5m 12 sec.

I managed to get rid of that problem by specifying in the rules the time when I needed indicator data.

It turned out that calculating only once a day (EA is working on days) I've reduced time of test calculation to 20 seconds and got time reserve for the test.

Now we can add more. Maybe it will help someone to get away from 5 min shock.

I'm going to go with the trend and make lots of profits.

 
That's right, Gep: counting every tick is clearly redundant, especially for a day-trading expert...
 
Almost any costly Expert Advisor, which is tested in tens of minutes (or hours), can easily be accelerated by a factor of 10-100.

Algorithmic optimization (acceleration effect is usually 1-2 orders of magnitude/10-100 times or more) is much more efficient than hardware acceleration (tens of percent effect or 2x at most). There were many times in this forum when traders complained that their EAs were tested for 24 hours and asked developers to "get the most out of their hardware, or use multi-core at full speed". Multi-core is certainly good. But if the author will rewrite his code more rationally, he will get an immediate speedup many times over, without involving developers of the terminal (which have long eaten a dog in optimizing their software and made it very efficient and speedy).
 
Mathemat:
And how many times faster this monster benchmarks against my old PIV Northwood 2.4 GHz + RAM 512 MB? That's actually why I asked about the benchmark, to avoid similar questions from others...


At least 1.5 times faster. But most likely even more, given the dual-core and 4Gb RAM.
 
That's understandable, Renat. I merely asked about the "modern Xeon, AMD X2" benchmark on which the automatic expert testing is going on. Forgive me for being stubborn, but the benchmark is a starting point from which any potential competitor can start from when sending their expert to the contest and without asking any clarifying questions. If, say, the corresponding MQ monster benchmark is 5000 and mine is 1000, then I can safely send an EA with a test time on my computer equal to 20 minutes (if there are no other obvious errors indicated earlier).

P.S. Question removed, thanks, Renat. All that remains is to find this test... I didn't even realise how good my CPU bought in early 2003 is :)
 
This is the Bench archive. I've put my own CPU test in there, it's a separate txt file.

The link is http://www.overclockers.ru/cgi-bin/files/download.cgi?file=450 (RAR archive), just in case.
Files:
cpu_ca.zip  429 kb
 
Mathemat:
P.S. The question was removed, thanks, Renat. The only thing I needed was to find the test... I even didn't realized, how good my CPU bought at the beginning of 2003 is :)
Well you can find it here http://www.overclockers.ru/softnews/17238.shtml And here is a variant of this test for Core Due 6600 overclocked by the system bus and by the CPU frequency by a factor of two
1234567891011
New comment