Genetic vs Random Cpu usage
26 replies
gentmat
8 years ago #113897
I am asking a pre-sales question to know what computer to buy for your application :
Of what i know ( and correct me if i am wrong ) its better to buy a xeon (Multi core) than a high clock speed CPU
So 24 Dual Core intel xeon (2.8 GHZ) is better than 8 cores (3.7 GHZ).
But this is true when using genetic algorith as the multi Thread will help evaluate all generation in different cores .
My question is the following :
I only use Random population ( Because genetic will lead to similar strategies whatever i do )
So does random population use multi core or only cpu high clock speed and cache .
Is the random strategies created each one on different core ? like the genetic ?
or if i have a 24 cores , the 24 cores will create 24 different strategies simultaneously
Thank you
gentmat
8 years ago #131454
Thank you for your experienced help . here and in pvt such a good forum with ppl like you .
The 4.1GHZ quadcore is AMD Phenom Black Edition 975 overclocked
The other is 2 xeon e5440s.
These 2 you can never compare , the xeons are “Intel” built and the 4.1 is a AMD . i am not saying AMD is bad but a 5.0 amd can not really beat a 2.5 intel core to core .
AMD doesnt even have hyper-threading .
The architecture of these cpus is totaly different . you need to compare the e5440 with lets say 2000 series or 4000 intel series
its not fair to compare it to a 5960x but thats what im planning to get in case i succeded with generating good strategies out of sq . But guess i ll wait sq4 to see its power and cpu consumption .
So a 5960x got 3.0 ghz …. overclock it with liquid cooling the min u ll get 8 cores running at 4.3 ghz . and max you ll get around 5 ghz if you have the lottery in a perfect ship (hopefully)
As xeons one would think of 2699 , a cpu with 18 cores running at around 2.4 , imagine a dual xeon 36 cores 72 threads (MAMA) … but each cpu cost 4000$ not worth it for me as i dont have such huge capital in forex to overcome such cost
Threshold
8 years ago #131456
E5440s don’t have hyper threading. I actually believe the AMD is superior in this comparison. Its a much newer chip. e5440s are old.
http://ark.intel.com/products/33082/Intel-Xeon-Processor-E5440-12M-Cache-2_83-GHz-1333-MHz-FSB
gentmat
8 years ago #131458
I was saying about the AMD 975 not the intel
gentmat
8 years ago #131459
e5440s will outperform even the newest fx AMD 5.0 which was about 2-3 years ago . AMD is stuck and i think ( my point of view ) after apple switched from AMD to intel . AMD
only surviving by some gamers that can not afford intel . I hope AMD succeed more because competition is good sign always but sadly intel owns all the way
Threshold
8 years ago #131460
Thanks for info.
gentmat
8 years ago #131461
NP,
Forgot to add if you guys need more speed try the following :
Lets say you only take on consideration sharp ratio > 1
add that to the filter both in oos and is
So the rubbish strategies doesnt go to the bank . When the strategy go to the bank the application use IO to write it in the database and that is physical storing as storing a .txt or .str file . so the hard disk will work and cpu makes huge load to write data in hdd or ssd ( ssd faster ) but still same .
So get rid of that . and a plus go to options and unselect checkbox store logs.
on h4 and with the recent post of sq tume optimisization using zulu rather than oracle im having on h4 around 130 000 generation per 10 mins almost 800 000 an hour
clonex / Ivan Hudec
8 years ago #131462
NP,
Forgot to add if you guys need more speed try the following :
Lets say you only take on consideration sharp ratio > 1
add that to the filter both in oos and isSo the rubbish strategies doesnt go to the bank . When the strategy go to the bank the application use IO to write it in the database and that is physical storing as storing a .txt or .str file . so the hard disk will work and cpu makes huge load to write data in hdd or ssd ( ssd faster ) but still same .
So get rid of that . and a plus go to options and unselect checkbox store logs.on h4 and with the recent post of sq tume optimisization using zulu rather than oracle im having on h4 around 130 000 generation per 10 mins almost 800 000 an hour
Excellent discussion gentlemans
Odoslané z SM-G900F pomocou Tapatalku
Threshold
8 years ago #131463
I don’t use genetic learning, just random generation, so it is pointless to have IS/OOS. Strategies that would have a bad period would automatically be filtered out. BTW sharp ratio isn’t a good measure. I know its an industry stand measurement but a lot of professionals criticize it with good reason. It can rank very good winning trend following trades with low Sharp scores, so it can be meaningless especially for a portfolio, or strategy that doesn’t have an even distribution with its standard deviation. So a strategy with a few very large winners will have infact a LOWER sharp ratio. This is often how trend following strategies operate. Small loss, small loss, small win, small loss, HUGE win = bad sharp (even though the strategy is a winner over the long term). Its designed for investors looking for smooth returns. Individual traders know big returns are never smooth.
SO thats why sharp is a poor measure and will actually filter out some good strategies.
I randomly generate from 4 different broker datas to avoid curve fitting.
Selected time frame only.
Minimum 500 trades, DD<30%, stagnation 0.8 (main and addition data)
Set 20 max strategies in databank and filter via custom weighted fitness (net profit, DD, std dev, stagnation, stability, profit factor) and let it run till I start getting high fitness scores and nice equity curves on each broker data.
gentmat
8 years ago #131464
Perfect explanation ! can you share a .xml settings such make it easier to put all these settings .
Threshold
8 years ago #131488
These were the last I used.
gentmat
8 years ago #131489
Thanks , i put it into test .