GSB BETA BUILDS

@brk, sorry for the late reply
Ive seen this before. Reboot your pc and try again. It might be due to low system resources. Default is 2500 systems. No great need to go above that
There also is a time setting in minutes, and max # compuations
check its like this.

continuing on crude oil research, and learning a lot.
If no improvements can be made, im very happy but sim still looking for minor refinements. New updates to GSB required for this.
There also will be a new GSB automation spread sheet with new features too
Hope to push out a GSB build late next week.
I have two new CL systems im live trading now
Shown is a hypothetical chart with 2.4 and $10 slippage per side

Quote: Originally posted by brk

I was just trying the CL settings on version 69.86 and tried copying them to Tradestation. I installed the GSB_ALL_FUNCTIONS_2025_0724.ELD. It seems the generated code calling GSB_AtrWilderFilter(...) function which generates an exception: Attempted to divide by zero (0). You should always check... Call Stack: GSB_ATRWilder.main()Input methodInput method GSB_HighestFFC.mai()GSB_Norm9.main()ELRte Do I have the correct ELD? Yes you do have the correct eld. Check you have a good number of years data (at least 3) and no holes in the data. If stuck email me and I will do remote support. You could also try the same system on different data. IE gc just to see if you get the error

Quote: Originally posted by admin

Quote: Originally posted by brk   I was just trying the CL settings on version 69.86 and tried copying them to Tradestation. I installed the GSB_ALL_FUNCTIONS_2025_0724.ELD. It seems the generated code calling GSB_AtrWilderFilter(...) function which generates an exception: Attempted to divide by zero (0). You should always check... Call Stack: GSB_ATRWilder.main()Input methodInput method GSB_HighestFFC.mai()GSB_Norm9.main()ELRte Do I have the correct ELD? Yes you do have the correct eld. Check you have a good number of years data (at least 3) and no holes in the data. If stuck email me and I will do remote support. You could also try the same system on different data. IE gc just to see if you get the error

macro location

Released yesterday. 69.90 build
Ability to export ts / mc code as plain text under saved systems.
TS/mc code will be in here
C:\GSB\Data\Scripts\....
also various of the existing stop entries, with offsets in ticks. My limited test to my surprise showed this was NOT improving results
Next build these features will be pushed so only the users in beta tester mode will have the stop offset orders

next build
Normalized closedlessOPend filter. (we have non normalized closeLessOpend filter only now)

out next week some time. build 70.15
New gold settings. These can be further refined and are likely still have remove for improvement, though im not yet sure how to improve them
systems gold 60 minutes, 24 hour data. Only a few systems in favouriteD, but systems are good any most are good since out of sample (May 1 2025) Whats unusual is i have never seen so many systems that all have negative correlation to each other. Note the first systems I built were session time ending with 1500 and 1700.
Shown is one GC system with the last year or so data.

improvements in performance metrics remarked out code, and the option to add this code with or without remarks. It now wont give performance metrics in print log for real time trading, and time sample in print log has seconds




fix error in value close, value open indicators. New elds needed for this and the gsb dummy script.
improvement in indicator imports from csv

This is for advanced GSB users. Ignore otherwise
Number of systems before restart. (NSR)
GSB has generations and populations in its genetic algorithms
conceptually there is an optimal range of these values, but it will vary per market and or bar interval
after the generations and populations have finished, GSB restarts with a new random seed of indicators / variables
It may take some time (seconds or low minutes) before systems are being built, and then you MIGHT get no new systems and number of systems just plateau's
beware, large population makes gsb much slower to build systems (last I tested years ago)
So you can use this to optimize GSB workers to get most speed. Note however speed of systems build varies greatly each time a new random seed is made.

released build 70.15
Has GC 60 minute swing systems settings
GC_20251001_60min-nth3_stp2k5_20250501OOS-40%InSample is designed for you to have a quick look, not build systems
GC_20251001_60min-nth3_stp2k5_20250501OOS-SysBuild is for system building. Do the usual get 10 best indicators, then second pass build systems.
Note you will only get a few systems in Favorites D
Macros are included. They differ from previous setups, and we are using nth3 instead of nth1
Enclosed are some .pa files from two systems I built.

also is out of sample results since 15/5/2025
and negative correlation of systems. Note the first 5 systems are day trading, the latter 2 swing (longer term) systems


Attachment: Login to view the details

Attachment: Login to view the details

no updates to GSB this week. Below is in the programmers job que to add into GSB
Note how the centerscaling and highestlowest2 were the previous normalization methods
signsquared scaling and sigmoid in ALL cases were better than these. for a few markets adaptive was also good.
This gave very significant in sample and out of sample improvement in system results
Im going to get all of these new methods added, then do testing to see what works.
note the synthetic data deliberately has high volatility period, followed by low volatility to show have the different normalization modes differ under these circumstances





Attachment: Login to view the details

Here are some initial stats on normalization modes
Whats clear is the original highestlowest2, and center scaling are no where near as good as the newer modes. Sigmoid and zscore are in the current build of GSB, and either are recommend to use as default (definitely not both of them at the same time)
Results in out of sample fitness F medium and the amount of systems in favorite D are my criteria

Whats the best normalization method varies per market, so im going to test again on Nasdaq long only systems

70.38 released. This has new dummy eld if you get a compile error saying looking for
dummyscript Value1 = GSB_Scripts_2025_11_10;

Next build multicharts files (.pla) as there are some differences that will cause TS eld to not compile.
Just for interest, now that I have many more normalization mode, im restesting them on known setups that work well on gold. Gold is just an amazing market right now.
Note again, how the traditional normalization modes centerscaling2 and highestlowest2 are way down the list compared to the new ones. zscore & sigmoid consistantly are among the best.
Im doing more tests on gold day trading now. This picture is for swing systems on 24x7 data
Hope to have a new build of GSB in a weeks time, but will know more closer to the time

Quote: Originally posted by admin

Next build multicharts files (.pla) as there are some differences that will cause TS eld to not compile. Just for interest, now that I have many more normalization mode, im restesting them on known setups that work well on gold. Gold is just an amazing market right now. Note again, how the traditional normalization modes centerscaling2 and highestlowest2 are way down the list compared to the new ones. zscore & sigmoid consistantly are among the best. Im doing more tests on gold day trading now. This picture is for swing systems on 24x7 data Hope to have a new build of GSB in a weeks time, but will know more closer to the time

Silver is just as good!

Quote: Originally posted by admin

70.58 released. Has pla file for mulitcharts, as some ts code will not work in MC. New ELD's too. Minor bug fixes, and code has been merged so same code can be used for .net 4.6 to 10, though likely some time before .10 will be released to users Spent the week testing normalization modes on numerous markets. Very cpu intensive as there are 60 normalization modes to test

Here is some of the normalization modes results
I will for users with beta tester mode, enable bollenger position and stochrsi smoothed. I have not yet tested how much better these are compared to the the current ones enabled.
not centerscaling had been the current default, then zscore50 / sigmoid
Will be out when i do the next release