SQ X updates
955 replies
Mark Fric
8 years ago #114916
Here’s where we can discuss SQ X updates nad news.
This topic was renamed from SQ4 Early preview. The old, no longer relevant posts are or will be deleted and archived.
The point is in clearing up our forum from old, non-relevant stuff.
Mark
StrategyQuant architect
chris
5 years ago #233965
I notice that the neural network still hasnt been incorporated in RC3, although in the custom projects the neural network trainer is there. Does this mean that the neural network is still going to be put into SQ4? Or are you planning on quietly dropping this feature?
Jojo
5 years ago #233976
Jojo
5 years ago #233977
Sorry, it was a temporary glitch. Now OK!!
Mark Fric
5 years ago #233985
I did a new folder installation and I can’t even open the program, it takes a long time starting strategyquant but it never opens it while with the previous version it would take just some seconds. This update looks awful for now…
Martin, what exactly do you mean that it never opens? Doesn’t it show Window with UI? Or it shows the window but UI is never loaded?
It would help if you’d submit the bug report with a screenshot, if you haven’t done it already
Finishing SQ – yes, we don’t want to add any new features until all the bugs are fixed. So features like Neural Network trainer or support for grid or Options to merge strategies when creating portfolios will be made as updates shortly after the final version is released.
Mark
StrategyQuant architect
yusung
5 years ago #233997
mikeyc
5 years ago #234007
Hi Mark,
An observation on all the SQ4 releases vs SQ3.
When I use SQ3 and set some simple filters on number of trades, profitability, drawdown etc, all the generated strategies tend to be rejected during generation based on those filters until a “good” one makes it into the list.
But with SQ4, almost all those rejected are due to “no trades”. It’s like the new engine generates mostly meaningless rubbish rules somehow.
I’m not really getting a sense that SQ4 works correctly internally, it somehow is broken at a fundamental level.
Regards,
Mike
mabi
5 years ago #234014
Dismissal stats. Been traveling and SQ4 RC2 been running for a week. Totally 67 mio strategies generated.
jmtc1230
5 years ago #234018
Mark – just like SQ3 can you add a window to the builder screen that actually shows the “last generation”?
I found this helpful to see what is actually happening DEEPER than the log file shows.
yourrodstaff
5 years ago #234027
I found the rc3 has very slow strategy generated, much slower than rc2, even considering the longer test data, I noticed low cpu usgae only 15% for my 32 cores.
tnickel
5 years ago #234028
hi yourrodstaff,
possible that you use different setting as in RC2.
On my computer 8/16 Cores, the cpu consumption is 99% (please post your setting).
You can open an error request with logfile and setting too.
thomas
https://monitortool.jimdofree.com/
mabi
5 years ago #234030
RC2 is about 36% faster then RC3 on my computors doing the same thing. RC2 works better then RC3 in my opinion and strategies found seems backtest indentically with Mt4 in most cases so il stay with RC2.. Noticed i can run 2 sessions aswell.
Jojo
5 years ago #234031
RC2 is about 36% faster then RC3 on my computors doing the same thing. RC2 works better then RC3 in my opinion and strategies found seems backtest indentically with Mt4 in most cases so il stay with RC2.. Noticed i can run 2 sessions aswell.
I have remained cautious in what I am doing – but RC3 is far more stable for me. NO crashes for time its been running. Previous versions crashed on me all the time. Third time lucky, may be!
jmtc1230
5 years ago #234034
Hello Mark. When I say this I believe I am speaking for the majority of our trading community. We will largely be using the software to find strategies that remain profitable in the worst case of all monte carlo testing. That really is the starting point for all profitable trading based on reviewing the workflow from many traders.
This brings me to a question. How good and how strong is the intelligence and logic of the software to find strategies that can perform well in this mc way? Can this be improved?
Does the software learn from strategies that fail, to then do more of what is profitable and achieves the desired results?
The reason I ask this question is because I have set up filtering purely based on monte carlo testing and a threshold limit, and after letting it run for days on end i’m getting zero or one or two strategies to look at.
I know this is kind of a nebulous question and general in nature but I’m just hoping to get your thoughts on this idea. Thank you.
Josh
mabi
5 years ago #234035
@jmtc1230, One way to use Genetic EVO which is the only intelligence available to generate only strategies that will pass MC is to set worst year performance to be above Zero. You see most strategies I have that have passed both MC and WFM have no or just 1 loosing year and that is after 2 years of constant generation on SQ3 with 160 threads. I guess it is supposedly because they are adaptive to the market and not curve fitted because it is very difficult to curve fit something to be profitable yearly for 13-15 years. That’s why I requested the option to have a worst year performance ranking option in SQ4. However I an not using it since crosschecks is now build in and there is also strategies with many loosing years that can pass MC and WF and I want all of them .. he he. Allthought that is impossible since the amount of possible combinations is in the trillions.
RNG
5 years ago #234036
ok this RC3 is totaly useless, you can’t save the retested strategies…