Bill, no wristdispect indented, but there are exceptions. If you can verify to your satisfaction that each parameter's infect is kindasortamaybe indespondent from every udder, you can have more than three parameters in a system. I have some very robust systems that have seven parameters, but there is a a lowerarchy of "strength" to them.
fastest optimization software ??
- Thread starter mcgene4xpro
- Start date
Bill, no wristdispect indented, but there are exceptions. If you can verify to your satisfaction that each parameter's infect is kindasortamaybe indespondent from every udder, you can have more than three parameters in a system. I have some very robust systems that have seven parameters, but there is a a lowerarchy of "strength" to them.
If they are independent this is a trivial problem. Make sure they are though. Sometimes the dependence occurs locally but enough to mess things up.
Two softwares are very fast, there is a comparison matrix on the tickzoom website... oddly enough, tickzoom does not compare their speed to AMIBroker and that might say a lot.. AMIBroker is compact enough to run entirely in the cache memory of the your processor, it never has to call any other system memory resources...
Assuming that we are only talking about processing time and that it make sense to optimize your problem.
1- Getting some more hardware is a costly but easy way to do it. But IMO, when you encounter an almost infinite number of runs, like here with 40 params it might not be a solution.
2- You can rent processing time on the GOOG or AMZN clouds. This one is also a costly solution.
3- You can rewrite your code in a compilable language, like C++.
4- You can use a profiler and look at the parts of the code that are time consuming and try to optimize them. Often you can skip a large part of code by doing some intermediate check.
5- You can rewrite your code with an optimisation algo. An optimization algo does not goe through all possible combinations. A good example is a genetic algorythm. You set some initial values to the parameters and then try some random combinations. When one combination is good you keep it and only modify a part of it. You are sort of keeping the good genes.
IMO for 40 params you need solutions 3 to 5 but 5 is the most important one. If you are not a programmer I think there are brokers that have genetic algos to solve the problems but I have no experience with them, I write it all by myself.
Good luck ;-)
Yea, i see they didnot compare with amibroker..
Thank you
Kindly help a senile old man out here, Bill. I have a question about the two parameters conomonon to all my systems. Is the opening time of the exchange you trade on indupendent of its closing time? That issue has always worried me. I have taken it on faith all these years, but maybe you have tested it. Thanks.
Thank you total_keops for your helpful answer. I think i will learn C to do it by myself on time. I will go with 2, rewrite with compiling language and optimize the code. At the same time, i am going to find away to exploit the speed of my hardware. However, the 40 parameters are not on the same level of importance in my humble point of view. i feel that my logic of my code still missing some hole(s) and optimization is trying to strengthen the other parameters to decrease this hole. btw, my code is showing massive profitability over long time and many trades ( some of my tests went from 50k to 1M in less than 10 days with +2000 trades on the 1 minute chart -90% quality). Unfortunately, it goes back to 120k by a massive trade. I am trying to maximize profitability for now and on time, i am going to exchange some profitability to get more stability and less risky trading. I am still in the beginning and not in rush
.Thank you very much. very helpful post.. However, programatically and mathematically i am striving in those two fields. However, i am still surviving..
Regarding the term " market conditions" , i have difficulty to understand it.. Very simple but very vague term for me
. I have another explanation regarding this mystery " market condition"I hope you know what out of sample, curve fitting, transaction fees, slippage and the like are. Also, make sure that your stops are hit b4 your limits in the backtest. When you test with time bars and not tick data you dont know what happened first within the bar.Quote from mcgene4xpro:
Thank you total_keops for your helpful answer. I think i will learn C to do it by myself on time. I will go with 2, rewrite with compiling language and optimize the code. At the same time, i am going to find away to exploit the speed of my hardware. However, the 40 parameters are not on the same level of importance in my humble point of view. i feel that my logic of my code still missing some hole(s) and optimization is trying to strengthen the other parameters to decrease this hole. btw, my code is showing massive profitability over long time and many trades ( some of my tests went from 50k to 1M in less than 10 days with +2000 trades on the 1 minute chart -90% quality). Unfortunately, it goes back to 120k by a massive trade. I am trying to maximize profitability for now and on time, i am going to exchange some profitability to get more stability and less risky trading. I am still in the beginning and not in rush
50k to 1M in less than 10 days looks to me like you are trading with some serious leverage or that you got some serious alpha skills. But it could also be some nasty men trying to get you in the car with some candies.
Have fun
To be honest, i still have this feeling that some nasty people are trying to get me in the car with some candies.. You are 100% write.. that is why i am trying to find the bottom of my trading system before going playing with the nasty guys.
your witty answer is amazing.. i appreciate it too much
