GSB Forums - Powered by XMB 1.9.11

my win 10 custom indicators is working fine, see picture below
If your at a total dead end, we can do teamviewer

Custom indicators

jptann - 12-7-2018 at 11:12 AM

Peter, I will try to generate a small custom file and send it onto you if it fails. I have reviewed the total "read only" issue with Windows 10 and have gotten into the weeds on Microsoft to make sure everything is OK.

Still no joy in custom indicator. Later today, I will attach a log file where I will substitute a standard function for one of my custom proprietary ones and then see what happens.

Thanks

admin - 12-7-2018 at 06:15 PM

Peter, I will try to generate a small custom file and send it onto you if it fails. I have reviewed the total "read only" issue with Windows 10 and have gotten into the weeds on Microsoft to make sure everything is OK.

Still no joy in custom indicator. Later today, I will attach a log file where I will substitute a standard function for one of my custom proprietary ones and then see what happens.

Thanks

First thing to clarify is can you write the CI file from within TS.
Worse case print to the print log, and put that in a file.

Custom indicators

jptann - 14-7-2018 at 11:59 AM

Peter:

I went back to TS alone and found that the custom indicator failed in TS for a FileAppend. Attached is the error event log from TS

I am going to substitute a standard indicator versus my proprietary indicator and see if that still gives the erro

jptann - 14-7-2018 at 12:56 PM

Peter: I took my custom function out of the attached TS custom file and it still fails on FileAppend in TS. There are a couple of issues:

1. When GSB CI generates its script, two lines appear in the // Settings Section that are not commented out. These are the DATA1 and DATA2 file. I commented these files out by adding // Without the // the file won't compile because a DATA1: or DATA2: can't start an instruction.
2. I'm pretty sure that the handling of these two statements are causing the CI to fail with a FileAppend.
3. I'm not sure what needs to be added to the Data1 and Data2 instruction that is under Settings to make the strategy compile. Commenting them out I'm pretty sure causes this error.

Thanks in advance

Attachment: Login to view the details

admin - 16-7-2018 at 01:36 AM

GSB 47.38 build. July 16th 2018
exe files only
Metrics now for each bar interval / market on nth test.
So you can build system on 29 30 31 min bars, and see the Nth OOS results on just 30 min etc. Not sure this is bug free in results.
Counters to identify if any workers are dropping systems.
Executes chime.wav and optional gmail.exe if the nth tests have completed.
This build will be superseded in the next week.
You need to put a sound file called chimes.wav in gsb folder

admin - 16-7-2018 at 04:52 AM

47.40 Fix in worker stats total, Input field for GSB icon

admin - 19-7-2018 at 04:01 AM

GSB 47.44. July 19. All users who use the cloud / manager should upgrade.
Fix GSB cloud issues.
Recent build changed cloud upload from 2 minutes to 10 seconds. This caused SQL server to drop trades and give exceptions for many users. Now we send systems after 1 min. Other fixes too.
Slippage and commission back in reports. For slippage & commission in fitness, run GSB in admin mode. I don't recommend different slippage and commission values for fitness and reports as its slows GSB down 10% or so
Full installer file.
One tip. GSBmanager, worker and standalone files are all identical. Its just the file name and the folder that determine function.
Another tip. Old workers wont typically run on new managers, but new workers typically run on old managers

admin - 20-7-2018 at 04:02 PM

47.44 has a count of sum of all systems on the workers, compared to manager unique systems count. Used to diagnose if you are dropping systems. This issue should be fixed in .44 and improved in .45 build
They should be identical, but subtract duplicate systems (small percentage), and subtract systems that come in and are discarded due to grace period. (under app settings workplace) (grace period 0 = disabled)
47.45 not released yet but has fix for some minor stats bugs.

admin - 25-7-2018 at 02:12 AM

47.46 will be out next. Testing build now. All users who use market verification feature should upgrade. I think the performance metrics were hard coded, so making changes to the defaults didnt work. (Thanks RWS for spotting)
Bug that nth stats B doesnt update. _ I think I was the only one with this issue.
We now have a timer that shows how long GSB has been waiting for workers to respond.

Petzy - 25-7-2018 at 05:21 AM

I run 47.44 still so maybe there is a way to do this in the new version?
How do I run a WFO when I genereate systems on 29,30 and 31 minute bars?

I get an error saying:
Walk-Forward is not allowed for multi-data system.

admin - 25-7-2018 at 05:23 AM

I run 47.44 still so maybe there is a way to do this in the new version?
How do I run a WFO when I genereate systems on 29,30 and 31 minute bars?

I get an error saying:
Walk-Forward is not allowed for multi-data system.

Good question.
You can make on 29 30 31, but only wf on 30 (or 29 or 31 etc)
use verification data of 30 min, then right click the system and verify. Then do wf

Carl - 25-7-2018 at 06:01 AM

Hi Peter,

I tried this, but get the same message as Petzy.

Build on ES 29,30,31
Verify on ES 30
Choose strategy
Walk Forward
Message: Walk-forward is not allowed for multi-data system

Am I doing something wrong?

Thanks

admin - 25-7-2018 at 04:22 PM

Quote: Originally posted by Carl

Hi Peter,

I tried this, but get the same message as Petzy.

Build on ES 29,30,31
Verify on ES 30
Choose strategy
Walk Forward
Message: Walk-forward is not allowed for multi-data system

Am I doing something wrong?

Thanks

I tried this, and got the same issue. Im not clear why as I was fairly sure it worked a week or so ago.
Bottom line is we will have WF of multi time frames soon. 2 or 3 builds away (1 week roughly) Im going to look into this issue regardless.

admin - 25-7-2018 at 04:38 PM

Next we will have stats where we can on say 29 30 31 min bars compare oos of say 29 to IS of 29. 30 to IS of 30. 31 to IS of 31.
Currently we compare oos of say 29 or 30 or 31 to the average IS of (20,30,31)

jptann - 27-7-2018 at 01:00 PM

Peter:

No down load files to .44 modifications

admin - 27-7-2018 at 03:40 PM

Peter:

No down load files to .44 modifications

You mean just the exe etc. Im getting too often trouble that less tech savy users are trying to run the exe files without all the other gsb files, and or run the old short cuts etc. Not sure what to do about it. Also users dont upgrade the workers, and find no systems build as often new manager wont work with old worker.
Old manager with new worker normally ok
Here is .48. Update to stats, but stats summary page not working yet

Attachment: Login to view the details

admin - 28-7-2018 at 04:29 AM

System build on 29,30,31 min bars and fitness is the average of all 3 bar intervals

In older build we
compare 29 OOS min with average in sample of (29,30,31)
compare 30 OOS min with average in sample of (29,30,31)
compare 31 OOS min with average in sample of (29,30,31)

now we also can
compare 29 OOS min with in sample of 29
compare 30 OOS min with in sample of 30
compare 31 OOS min with in sample of 31
Enjoy

admin - 1-8-2018 at 12:00 AM

build 48.01 has not passed quality control. We have found that some virus checkers are saying that the files GSB manager
loads into windows temp folder are a virus. (Strange as there is no exe etc in them)
In 48.01 on-wards the location will be changed to be a temp folder under GSB. Its not a good idea to exclude windows temp folder from anti virus, but GSB folder to exclude is fine.

admin - 5-8-2018 at 08:39 PM

48.07 out hopefull in a fe days. A number of things you need to know.
CloseD-Close is not a good idea unless you factor in big point value - when used for market validation on markets that differ a lot in points. ie dow vs ES
The typical value was about $800
This now means ALL contracts must be in the contracts list. This includes cash indices.
However im thinking this through. I think we need to use bigpointvalue of the matching futures contract, on the cash indices.
Otherwise if we use close(d)-close on data2, where data2 is a cash indices, market validation from ES to dow still wont work well.
Shown in the first picture is a way to clone one contract to others.

Icons from manager now propagate to workers.
There are many bug fixes in 48.07

There is also a test for workers loosing systems if GSB is in admin mode.
There are 557 systems that should make there way to the manager in time. But 671-557 are duplicates so will be discarded.
Also systems that come after gsb has stopped (+ grace period under app settings/ workplace) will also be discarded
There is a time lag for the wp dn (worker downloaded) to reach the manager

There is also a metric to tell you iterations per system made.
This is 275148/501 - 549. Example is CL 29,30,31
If we had just cl 30, the ratio would be about 22

This also means there is a new TS GSB function. "GSB_CloseLessPrevCloseDBpv"
Should not cause any issues with older code. Note however that the closeD functions used session as of a few weeks ago. This was purely to fix a few users on time zones like Australia. Turns out closed(1) <>closesession(1,1) in all situations, we we are removing closesession(1,1) for now

cyrus68 - 6-8-2018 at 11:58 PM

As I understand it, if we want to verify a market on other markets. For example, ES verified on YM and NQ (where SPX, IDX or other symbols are included as secondary data). Then all the symbols need to be specified in the contracts list.

However, if a given market is verified on different time frames, this is not needed. For example, using ES 25min on data2, data3; ES 30min on data2, data3; ES 35min on data2, data3.

admin - 7-8-2018 at 12:08 AM

As I understand it, if we want to verify a market on other markets. For example, ES verified on YM and NQ (where SPX, IDX or other symbols are included as secondary data). Then all the symbols need to be specified in the contracts list.

However, if a given market is verified on different time frames, this is not needed. For example, using ES 25min on data2, data3; ES 30min on data2, data3; ES 35min on data2, data3.

correct on all points.

admin - 7-8-2018 at 01:49 AM

48.07 still has some critical bugs so cant be released. One of the bugs is the wf parameter results dont match TS. Its out by a mile.
Enclosed is a CL system that was built on all bars from 25_35 min. Its oos 2018.1.1
Session times 900 to 1430. The csv data exported will end at 1330 as for me local time <> exchange time
This system will pass a wf but this is busted in current gsb build.

Attachment: Login to view the details

admin - 12-8-2018 at 06:31 PM

New GSB code needed for next version. It should not cause issues with any older GSB code, with the exception of GSB_closeDminus & GSB_CloseOverPrevCloseD functions. These had a fix for no trades in time zones like Australia. This fix is being rolled back as
closesession(1,1) doesnt always = CloseD(1)
There is also a fix in these functions for @RB which can hit a price of zero in 2007. This caused GSB<>TS. Why RB hit zero price is due to back adjusted rollover calculations used in TS.
The free GSbsys1ES has a exit added and tighter stop.
Attachment: Login to view the details

version .12 (still testing has these updates)
1.0.48.12 / 2018-08-13:
Moved saved price data settings from "PriceData.txt" to "Price Data.txt".
Improved LoadPrice()/LoadPRiceFiles() logic to be faster and use less memory.
Fixed cloud WF on manager's side.
Fixed too much, unreleased memory on Load system.
Updated Load systems logic to discard exsiting systems.
Improved GSB logic for better cloud WF management based on LastWorkerSearch/DateTime rather than a random number.
Auto cleanup unassigned WF after 4 hours instead of 24 hours.
Auto remove duplicate bars to overcome the TS exporter script limitations.
Fixed CloseLessPrevCloseDBpv, by using * instead of /.
Added CloseLessPrevCloseDBpv disabled by default, until testing completed.
Updated ResourcesMonitor.WorkplaceOptimizationStatus to show "initializing.." until all files uploaded, then "searching.." while contacting workers.
Uploading and downloading files in multi-threads.
AppSettings.WorkplaceWorkerGroupPercentage default updated from 10% to 100%.
OptimizationSettings.MainPriceData added and partially implemented.
Fixed OptimizationSettings.CheckContractsMatch(), contracts clone and question message window in the background.
Added nth info to GSB and TS debug logs.
Fixed nth day causing GSB <> TS mismatch (not well tested).
Fixed cloud WF not showing on systems and unique-systems grid.
Globals.PriceFile and Globals.Prices implemented.
Updated verify logic to allow prices without equal number of files.
Updated verify logic to discard duplicate prices that's in opt. price data.
Updated prices window to preserve last open folder.
Fixed the frequent loaded/global contract mismatch window by auto updating contracts every time the prices window is opened.
Fixed the frequent LogException() exception.
Updated new workplaces and walk-forwards to not accept workers until all files uploaded.
Minor GUI updates here and there.

admin - 15-8-2018 at 11:22 PM

We still have one hard to find bug before release of 48.14
48.14 is still being tested by one trial user and myself.
Note we now have (optional) fields for price data.
Opt Price data is used for all, unless you use the other live, verify, WF fields.
Shown here I built a system on 29,30,31 min bars.
will trade it live on 30 min bars.
Verified the system using all bars from 25 to 35 except 29,30,31. (No need to use 29,30,31) as it been done already when the system was built
WF. I am Wf this on 29,30,31 min bars

There also is now an option to deselect all data. You might want to do this say on WF, where when its deselected- it reverts back to opt price data.

Petzy - 16-8-2018 at 02:46 AM

I am trying the 48.12 version and it is a very nice version with the different price data choices.

I am trying different files to achieve “true oos”. I have 3 data files with the end date of 2015, 2016 and 2017. I have the 2015 for calculating the strategies and then I verify with the 2016 files. I am saving the 2017 to use the “Override Original Settings. That way I should be able to develop systems and then test them against unseen data. (The very last test is with TradeStation with data up until current date)

I think I found a bug with the price data though. It doesn’t save the price information correctly in price data.txt. When I quit gsb and start again the data is lost. (I edited the file manually in notepad so I am good. But just so you know.
At one point I got the old problem with information that changes when I click on the strategy in gsb manager. I think this error is an old one that was fixed before. I have only got this problem once so I cant reproduce it at the moment

admin - 16-8-2018 at 05:44 AM

I am trying the 48.12 version and it is a very nice version with the different price data choices.

I am trying different files to achieve “true oos”. I have 3 data files with the end date of 2015, 2016 and 2017. I have the 2015 for calculating the strategies and then I verify with the 2016 files. I am saving the 2017 to use the “Override Original Settings. That way I should be able to develop systems and then test them against unseen data. (The very last test is with TradeStation with data up until current date)

I think I found a bug with the price data though. It doesn’t save the price information correctly in price data.txt. When I quit gsb and start again the data is lost. (I edited the file manually in notepad so I am good. But just so you know.
At one point I got the old problem with information that changes when I click on the strategy in gsb manager. I think this error is an old one that was fixed before. I have only got this problem once so I cant reproduce it at the moment

Had someone with I think it was bit defender. Didnt allow any writes in c:\gsb.
Thats just my wild guess, as I have had no issues saving data.
We can do Team-viewer tomorrow if you have issues with 48.14

admin - 19-8-2018 at 04:00 PM

GSB 48.16. August 20. bug fixes, inc fix for workers not starting. (cause was identical file contents with separate file name)
SF closedbpv ,and close/closed (both not normalized)
My limited tested shows non normalized close/closed does NOT work well. Option to right click and delete data in data manager
YOU MUST HAVE ALL DATA 2 ENTERED IN CONTRACTS LIST. Note also your groups of data (ie 29,30,31) are not saved in contracts, but in opt settings. (This might change later)
Other minor updates
To use this version, put the files in C:\GSB\GSB (Managers) & C:\GSB\GSB (Workers) and run these files instead of .44 version.
There is no installer file with this build. You need 47.44 build installed first.
See top of thread fix exe file.
You also need new ELD. In it is closeLessCloseDBPV function, GSB_Scripts_2018_08_05, update to GSBsys1es (new exits, tighter stop), bug fix for GSB_supersmoother. None of these updates should cause issues with existing GSB systems.

Bruce - 19-8-2018 at 08:05 PM

Do you have an updated contract list that could be shared that has all the required data for SPX, IDX, etc?

admin - 19-8-2018 at 08:11 PM

Quote: Originally posted by TradingRails

Do you have an updated contract list that could be shared that has all the required data for SPX, IDX, etc?

Thats a good request.
I will work on that in time for the next installer package.
Its very fast to do, in that you can right click and existing symbol - clone,and add multiple new symbols.
Busy today on the next video

admin - 26-8-2018 at 08:12 PM

GSB 48.23. August 27. bug fixes and new features.
Right click to apply WF parameters to the systems. Known intermittent bug. WF to the cloud has nth set to no trade, when it should be all. This is not the case when WF to manager is done.

To use this version, put the files in C:\GSB\GSB (Managers) & C:\GSB\GSB (Workers) and run these files instead of .44 version.
There is no installer file with this build. You need 47.44 build installed first.
New GSB Functions were not in the last ELD. Missing was the needed but dummy function GSB_Scripts_2018_08_05;
Fixed missing manager's version (note: workers update is required).
Fixed cloud WF Nth.
Update drop-down controls to preserve last selected item on selecting a new system in the grid.
Fixed PA's WF trades export/save.
Fixed support exception caused by price files not in the root "Price Data" folder.
Fixed support missing "Price Data.txt" file.

1.0.48.22 / 2018-08-22:
Updated "Use WF Params" and "Override Original Data" to auto backtest all selected systems.
Fixed Stats grid cross-sign exception (in v48.21).
Fixed multi-WF start exception.
Updated Utils.InvokeIfRequired() to log exceptions before throwing them.
Set support's default to include screenshot of the underlying window.
Fixed wrong MAC address on machines having VirtualBox installed.
Set AppSettings and OptimizationSettings FullPath property internal/hidden and substituted its usage by FileName

admin - 8-9-2018 at 04:49 AM

49.03 should be out Tuesday or earlier
Data manger now has 1 minute bars. bug fixes and small tweaks

admin - 10-9-2018 at 04:27 PM

GSB 49.05 Sep 11
support for all bars derived from 1 minute bars. Bug fixes, multiple export to PA files. To use this version, put the files in C:\GSB\GSB (Managers) & C:\GSB\GSB (Workers) and run these files instead of .44 version.
There is no installer file with this build. You need 47.44 build installed first.

New system

jptann - 14-9-2018 at 10:55 AM

Peter:

I am running 49.05 and use the ES29M and the SPX 29M back till 2000. During the new releases, I stopped obtaining unique systems. Nothing shows up. A couple of comments. I have been using the opt price data for about two years now so nothing wrong here. Do I need to generate the time frame from 1m bars????

Under the Live Price Data (which I am not using) it still has a data file and I can't clear that field. Shouldn't I have this field as blank. Finally, I have the verification on for 27, 28, 30, 31, and 32M

By the way, I have not changed any of my performance specs to identify unique systems.

I am going to start a run with a file that has no restrictions for uniques systems. Lets see what happens with that!!!

admin - 14-9-2018 at 03:54 PM

Peter:

I am running 49.05 and use the ES29M and the SPX 29M back till 2000. During the new releases, I stopped obtaining unique systems. Nothing shows up. A couple of comments. I have been using the opt price data for about two years now so nothing wrong here. Do I need to generate the time frame from 1m bars????

Under the Live Price Data (which I am not using) it still has a data file and I can't clear that field. Shouldn't I have this field as blank. Finally, I have the verification on for 27, 28, 30, 31, and 32M

By the way, I have not changed any of my performance specs to identify unique systems.

I am going to start a run with a file that has no restrictions for uniques systems. Lets see what happens with that!!!

email me team viewer details and I will look. One clue is is fitness appearing in the top right corner. If its blank or zero, internally no systems are produced. You can also right click on a worker, or standalone the top/diag field and look at the top systems GSB produces.

admin - 17-9-2018 at 05:13 PM

GSB 49.08 18 Sept
Much improved ram usage for NTH tests.
GSB format system encrypted at the end of TS code. The can be reverse engineered to get the GSB format code if you loose the GSB format code.
Has % progress for nth / dates change. Useful for when your doing high number of systems. Big improvement to ram usage in nth tests.
All GSB code will expire at the end of the month. I expect a number of new features still to come in the next week.
There is no installer file with this build. You need 47.44 build installed first.

admin - 19-9-2018 at 01:41 AM

GSB 49.09 19 Sept 2018
CloseDBPV now in secondary filters.
User task gui on right hand side now includes nth, dates, and verification progress. On all GSB versions do not do nth and date changes at the same time. Results will be wrong. Wait for each to finish. Other bug fixes. Help file included under help.
Code expires mid Jan 2019
No installer file.

admin - 21-9-2018 at 06:15 PM

48.12 is under testing. There is a new field nth day period. It used to be hard coded to 1.
Now we can use 5 days, or 200 days etc where we have our trade / no trade windows.
The reason is for non MOC nth tests, the nth logic is a little flawed.
Nth will exit all trades when nth goes to no trade. Thats a issue.
Also it may take a trade late that would have been taken in the time when it was no trade.
This covers up this issue. A better fix will happen, probably after exits and secondary filters are done
The field will only appear when GSB is in admin mode.
https://trademaid.info/forum/viewthread.php?tid=92

cyrus68 - 22-9-2018 at 02:17 AM

I don’t use swing strategies currently. But clearly the Nth day mode would be flawed. I don’t see how using the Nth day Period solves the problem.

As for MOC strategies, unless I’m wrong, the Nth day mode should still be viable.

admin - 22-9-2018 at 02:20 AM

I don’t use swing strategies currently. But clearly the Nth day mode would be flawed. I don’t see how using the Nth day Period solves the problem.

As for MOC strategies, unless I’m wrong, the Nth day mode should still be viable.

This covers up the problem. If nth is is 1 month periods, the issue happens less often. Its very fixable and a perfect solution can be done to. Just not sure if it will be very soon or after secondary filters and exits.

nth for moc is perfect.

admin - 24-9-2018 at 12:50 AM

I posted 49.12 trial version in the main forum (not private)
It has installer file but might overwrite your settings. No need to install it if your on 49.12 without the installer file.
Note future GSB build needs to have its own bigpointvalue GSB function, as GSB will work best when BPV futures = BPV cash indices.

admin - 30-9-2018 at 07:45 PM

I reported that nth day period was buggy. This turns out to not be the case. It works fine.
Next build has option too do nth, then toggle dates, then execute.
the current mode is
nth to all, (execute instantly & wait), dates to no trade (execute instantly & wait)

This makes it much faster if you go from nth=trade, dates = trade to dates = dont trade, nth = all.
It means we do one lot of processing, not two.

Normalized nth is done,
It means if nth = say 2, the nth stats takes into account the results are degraded due to less trading days.
If you dont do this, it means the number of trades, NP and fitness are all degraded by extra 1/3

nth>1 is needed for say copper, were we struggle to get 100 trades with nth=1. It should also give a little better market degradation stats as there is more data in sample.

cotila1 - 1-10-2018 at 01:12 AM

all those things already in the next build? which should be 49.10?

I reported that nth day period was buggy. This turns out to not be the case. It works fine.
Next build has option too do nth, then toggle dates, then execute.
the current mode is
nth to all, (execute instantly & wait), dates to no trade (execute instantly & wait)

This makes it much faster if you go from nth=trade, dates = trade to dates = dont trade, nth = all.
It means we do one lot of processing, not two.

Normalized nth is done,
It means if nth = say 2, the nth stats takes into account the results are degraded due to less trading days.
If you dont do this, it means the number of trades, NP and fitness are all degraded by extra 1/3

nth>1 is needed for say copper, were we struggle to get 100 trades with nth=1. It should also give a little better market degradation stats as there is more data in sample.

admin - 1-10-2018 at 02:36 AM

Quote: Originally posted by cotila1

all those things already in the next build? which should be 49.10?

All in 49.20 - has been release to a few testers. It has these features and custom dates for training, test, validation.

admin - 1-10-2018 at 01:00 PM

GSB 49.20 Oct 2 2018
Has nth normalization, improvements to nth / dates job que. Statistics are normalized to take into account how many days are traded in stats a vs stats b. Training, test and validation now can have unlimited custom dates.
No installer file. Cope exe into your current GSB folders.

admin - 4-10-2018 at 07:02 PM

Under testing now GSB 49.25. It has bigpoint value from GSB contracts, not TS BPV
This might cause GSB<>TS if you have used a cash indices for data2 and closedBPV function.
If it concerns you, then make new cash contracts as shown in the png.
49.25 has one bug that must be fixed before release.
New ts code also needed.
GSB_CloseLessPrevCloseDBpv2, GSB_Scripts_2018_10_04
The logic for this is if GSB genetically switches data2 from $spx.x to $indu, if bpv=1, it will work badly unless the closeD offset also changes.
Making bpv on the cash indices the same as the futures will fix this.

admin - 5-10-2018 at 12:18 AM

GSB 49.28 Oct 2 2018
Bug fixes and ClosedBPV from GSB contracts. See note in post above.
This build is not yet well tested.
See top post for file(s)

cyrus68 - 5-10-2018 at 03:32 AM

There is a need for clarification on the BPV issues.

First, the big point value on TRADED contracts in GSB are set to be the same as on TS. So what's the issue?

Second, you posit that all indices entered as secondary data (e.g. SPX INDU as 2nd data for ES) should have the same BPV and ticks as the traded contracts. In this case, ES and YM. If so, we can create new ones in the contracts list. However, it is not clear how this is implemented in TS, where the indices have different point values from the ones in GSB? A test should be able to verify if the new scripts work and if TS<>GSB.

Third, the parameter settings in the PNG - is that only for closedBPV or for the other filter types as well? Does it apply to all futures and stocks or not?

Fourth, these parameter settings would apply to the normalised SF. What about the non-normalised ones?

admin - 5-10-2018 at 05:02 AM

There is a need for clarification on the BPV issues.

First, the big point value on TRADED contracts in GSB are set to be the same as on TS. So what's the issue?

Second, you posit that all indices entered as secondary data (e.g. SPX INDU as 2nd data for ES) should have the same BPV and ticks as the traded contracts. In this case, ES and YM. If so, we can create new ones in the contracts list. However, it is not clear how this is implemented in TS, where the indices have different point values from the ones in GSB? A test should be able to verify if the new scripts work and if TS<>GSB.

Third, the parameter settings in the PNG - is that only for closedBPV or for the other filter types as well? Does it apply to all futures and stocks or not?

Fourth, these parameter settings would apply to the normalised SF. What about the non-normalised ones?

1) lets pretend $spx.x is the best indices for closeD
GSB is on es for closed with say $800 secondary filter parameter
if gsb genetically moves to $spx (from ES), at 1$ a point, its 800 points and the system wont trade. So GSB genetically is unlikely to find that $spx is better than
ES. The switch from ES to $spx is easy if we have $spx at $50 a point.
same issue switching from ES to YM (dow)

2) GSB will use its own bpv value derived from its contracts table. It wont use TS bpv

3) if stock were used with futures, then you might not want bpv=1, but this isnt common. Its not critical to get right, but will work better if you do get it right.
This is for BPV only, and wont affect anything else.

4) If its normalized, the parameters wont matter. For non normalized, es emd ym er are all about $800 roughly. so 0 to 1500 step 50 would be a good default

cyrus68 - 8-10-2018 at 07:28 AM

The explanation is useful, but questions remain regarding the use of the secondary filter. It is critical to get it right because of the out-sized impact of the secondary filter on results. I'm still not sure of the correct setup for using the secondary filter.

First case. Data1=futures and data2=futures, where the data2 futures may have very different point values, relative to data1 and each other. For example, data1=ES, data2=EC data3=HG. How are we supposed to set this up for closedBpv?

Second case. Data1=stocks and data2=futures, where the data2 futures may have very different point values, relative to data1 and each other. For example, data1=AAPL, data2=SF data3=NQ. How are we supposed to set this up in the secondary filter?

admin - 8-10-2018 at 02:29 PM

The explanation is useful, but questions remain regarding the use of the secondary filter. It is critical to get it right because of the out-sized impact of the secondary filter on results. I'm still not sure of the correct setup for using the secondary filter.

First case. Data1=futures and data2=futures, where the data2 futures may have very different point values, relative to data1 and each other. For example, data1=ES, data2=EC data3=HG. How are we supposed to set this up for closedBpv?

Second case. Data1=stocks and data2=futures, where the data2 futures may have very different point values, relative to data1 and each other. For example, data1=AAPL, data2=SF data3=NQ. How are we supposed to set this up in the secondary filter?
It doesnt matter if you get this wrong, as its not critical and just means your not at top efficiency. Ideally you want the data1 and data2 to have similar values relative to their movement and volatility range.

cyrus68 - 8-10-2018 at 10:08 PM

"Ideally you want the data1 and data2 to have similar values relative to their movement and volatility range."

I thought that this issue is addressed in GSB by normalising the data streams.

The problem that I'm focusing on, is the treatment of point values in the secondary filter. If the difference in point values between ES and SPX is a problem, then the difference in point values between AAPL and SF is an even bigger problem. You are saying that it's just too bad and we should ignore it.

Another thing that I noticed in version 48.28 is that the secondary filters are now only available under the indicators menu. Presumably this means that only normalised filters are now available, though this is not entirely clear.

admin - 8-10-2018 at 11:17 PM

" If the difference in point values between ES and SPX is a problem". No its not a problem, its just an area than can and has been improved on a bit. Im not saying its too bad - im saying its not very important and that you can ignore it. There are much more critical issues, which I will put in the forum soon.
None of thep point values is an issue when things are normalized, but at times Secondary filter works best not normalized.
Im not aware of any changes in 49.28 with SF. They have been in that format for a long time. What did change is secondary filter has non normalized BPV and c/closed(1) added.

Please make a post where everything is explained regarding the new point value, secondary filter and changes in parameters

Petzy - 9-10-2018 at 01:18 AM

The new changes in GSB regarding the point value, secondary filter and the change in parameter is getting way to complicated for me. I am completely lost now. There are some scattered explanations in the forum but it is hard for me to understand how it all ties together.

It would be very appreciated if there was one post or one document with a step-by-step instruction explaining how to setup GSB.

At the moment I don’t really know how to use GSB. The results in GSB is no longer the same as in TradeStation.

admin - 9-10-2018 at 01:22 AM

The new changes in GSB regarding the point value, secondary filter and the change in parameter is getting way to complicated for me. I am completely lost now. There are some scattered explanations in the forum but it is hard for me to understand how it all ties together.

It would be very appreciated if there was one post or one document with a step-by-step instruction explaining how to setup GSB.

At the moment I don’t really know how to use GSB. The results in GSB is no longer the same as in TradeStation.

IF gsb<>ts , send me team viewer details. There is nothing in new builds to cause this that Im aware of. Docs are getting there, but got some time to go. The help file is updated regularly. The new BPV closeD is not critical to use, but slightly better in some circumstances.

cyrus68 - 9-10-2018 at 02:52 AM

I agree with Petzy. GSB is looking increasingly complicated, introducing new mysteries that I can't fathom - let alone the issue of new bugs.

The whole issue of using WF in the context of multiple datasets. What is being walkforwarded and what parameters are being used?
How to view the results on the graph, replete with lots of other lines.

admin - 9-10-2018 at 02:59 AM

I agree with Petzy. GSB is looking increasingly complicated, introducing new mysteries that I can't fathom - let alone the issue of new bugs.

The whole issue of using WF in the context of multiple datasets. What is being walkforwarded and what parameters are being used?
How to view the results on the graph, replete with lots of other lines.

Well docs are still a work in progress, and we are all learning the best ways to use GSB. GSB advances at such a rapid rate docs lags. Other products have current docs, but little product development. I would rather have product development.
I would rephrase that to say GSB is increasing powerful rather than complex. There are more features you dont need to use, but are better off learning. You are in the beta forum so expect bugs and delay docs.. Otherwise stick to the release builds. Most new versions have bug fixes, thought they are not always reported.
You just need be patient as there is only so much that can be done on a week by the GSB support staff.

cyrus68 - 10-10-2018 at 12:53 AM

So, to confirm, are you saying that if the normalised versions of the secondary filter are used, then differences in point values do not matter?

Here's one of the mysteries. Previously, we were told that the secondary filters in the GUI drop-down list were non-normalised and those in the indicator list were normalised. Well, in version 28 the choices in the GUI are Disabled, Enabled and Walkforward only. I still haven't figured out what the last one does.

So, where are the non-normalised SFs? Are the ones in the indicator list still normalised?

admin - 10-10-2018 at 01:00 AM

So, to confirm, are you saying that if the normalised versions of the secondary filter are used, then differences in point values do not matter?

Here's one of the mysteries. Previously, we were told that the secondary filters in the GUI drop-down list were non-normalised and those in the indicator list were normalised. Well, in version 28 the choices in the GUI are Disabled, Enabled and Walkforward only. I still haven't figured out what the last one does.

So, where are the non-normalised SFs? Are the ones in the indicator list still normalised?

MY REPLY IN CAPS
So, to confirm, are you saying that if the normalised versions of the secondary filter are used, then differences in point values do not matter?
CORRECT
I BELIEVE EVERYTHING IN GA IS NORMALIZED.
WF SF WOULD BE BUILD SYSTEM WITH NO SF, BUT WHEN YOU WF THE SYSTESM ADD IT IN. I DONT KNOW NAYONE WHO HAS DONE THIS.
NON NORMALIZED ARE HERE. ITS POSSBILE IF YOUR NOT IN GSB ADMIN MODE THAT BPV IS OGNE IN .28 VERSION. ITS BACK IN .30 VERSION

cyrus68 - 10-10-2018 at 02:35 AM

Here's what I see.

I don't see the value in building a system with no SF and then then add it in when running WF.

GUI SF.bmp - 271kB

admin - 10-10-2018 at 03:46 AM

"I don't see the value in building a system with no SF and then then add it in when running WF." Well thats fine, Ive never done it but i could be done
here is .31. Has a fix for oyur wf issue and the sf bpv should be back. I have not tested this build at all. Time for me to stop for the day.
this build will not talk to any other versions

Attachment: Login to view the details

Carl - 10-10-2018 at 07:33 AM

Hope to clarify the closelessprevclosedBPV enigma...

BPV = bigpointvalue = futures contract size.

Suppose you are using closelessprevclosed as the secondary filter (non-normalized).
And suppose you are testing on ES. You likely get the best results if closelessprevclosed is somewhere between 14 to 20. This means a 14 to 20 points price change calculated from yesterday’s close.
If you verify this strategy on YM, the results will be horrible. Because you are using a price change of say 16 points on YM. This doesn’t work. On YM you need a closelessprevclosed in the range of 150-200 points.

So 16 points on ES corresponds to 16*50 = 800 USD price change
And 160 points on YM corresponds to 160*5 = 800 USD price change

The 800 USD is the closelessprevclosedBPV and can be used on YM and on ES. So the closelessprevclosedBPV (in USD) is the more generic or universal version of the closelessprevclosed (in points).

admin - 10-10-2018 at 02:17 PM

Well described Carl. Wonder who thought of such a good idea

cyrus68 - 11-10-2018 at 02:39 AM

Carl has decidedly good language and explanation skills. The example that is utilised, considers parameter settings when YM is used for verification purposes.

I’m still unclear on settings and contract listing specs when we have no interest in verifying on other contracts but wish to include them as secondary data. In Carl’s example, let’s have data1=ES data2=SPX data3=YM and data4=EC.

According to Peter’s previous explanations, if we use normalised versions of close-close, close/close and closeBpv, then there is no need to change the contract information for the secondary data or worry about SF parameter settings.

However, he has also stated that when GSB is considering the various data-streams for inclusion/exclusion during system building, it would find it difficult to handle the difference in point value between ES and SPX. So, a new SPX contract must be created, with point values equivalent to ES.

In the normalised versions of the secondary filter, do we need to define a special contract for SPX? And what do we do about YM and EC? Point values, parameter settings? And for the non-normalised versions?

admin - 11-10-2018 at 04:00 AM

Carl has decidedly good language and explanation skills. The example that is utilised, considers parameter settings when YM is used for verification purposes.

I’m still unclear on settings and contract listing specs when we have no interest in verifying on other contracts but wish to include them as secondary data. In Carl’s example, let’s have data1=ES data2=SPX data3=YM and data4=EC.

According to Peter’s previous explanations, if we use normalised versions of close-close, close/close and closeBpv, then there is no need to change the contract information for the secondary data or worry about SF parameter settings.

However, he has also stated that when GSB is considering the various data-streams for inclusion/exclusion during system building, it would find it difficult to handle the difference in point value between ES and SPX. So, a new SPX contract must be created, with point values equivalent to ES.

In the normalised versions of the secondary filter, do we need to define a special contract for SPX? And what do we do about YM and EC? Point values, parameter settings? And for the non-normalised versions?

anything normalized, point value doesnt matter.
As for non normalized this has been discussed already to the max extent it needs to be.

admin - 14-10-2018 at 05:38 PM

GSB 49.35 Oct 15 2018
Additional debug mode if TS <>GSB
A few issues that cause TS<>GSB.
Some issues were holidays, (short session time) stops price out by a tick, TS not handling numbers very close to zero.
timed minutes exit in TS <> GSB

admin - 17-10-2018 at 01:24 AM

In the pipeline, 49.40
option of timed exits after x bars, rather than x minutes
We not have internal gmail support to send email once jobs completed.
a few bug fixes.

Im hoping for market degradation figure on each system.
ie we make systems on 29,30,31 min bars, and validation on 20,40 25,25. The system degrades x% between the two sets of bars.
Then we can the individual systems by degradation.

waldocktrades - 17-10-2018 at 08:03 AM

Will the "x-bars" exit be optimizable or, pre-set in the GUI?
I've been working on an indicator that tracks a market's day by the time period in a manner similar to seasonality. It looks like it will be a useful tool in predicting average point differentials from time "a" to time "b."
Thanks for all of the support, thus far.
Andy

admin - 17-10-2018 at 04:32 PM

Quote: Originally posted by waldocktrades

Will the "x-bars" exit be optimizable or, pre-set in the GUI?
I've been working on an indicator that tracks a market's day by the time period in a manner similar to seasonality. It looks like it will be a useful tool in predicting average point differentials from time "a" to time "b."
Thanks for all of the support, thus far.
Andy

In the next build it will be fixed.
What your trying to do is very good, and we should add into GSB.
Im just trying to think this through.
1) genetically build systems and optimize times of the day. //to prone to curve fit.
2) build 1000's of systems and look at the spread of times used, then build systems with the chosen times as fixed variables.
Andrea Unger is big into market bias's, so your in good company.

Im open to more ideas how this might work.

boothy - 17-10-2018 at 07:39 PM

Something I heard on a podcast the other day might be useful with x-bars exit and also with general system validation is E-Ratio

http://bettersystemtrader.com/079-strategy-validation-dave-b...

https://www.buildalpha.com/eratio/

linked is the podcast and an article on how to use and calculate it.

Thoughts??

admin - 17-10-2018 at 07:50 PM

Something I heard on a podcast the other day might be useful with x-bars exit and also with general system validation is E-Ratio

http://bettersystemtrader.com/079-strategy-validation-dave-b...

https://www.buildalpha.com/eratio/

linked is the podcast and an article on how to use and calculate it.

Thoughts??

So what we would do is have eratio for each system, and then do nth stats on the good eratio and compare with the bad eratio? THat would show if the concept works or not. Im open to other ideas?

boothy - 17-10-2018 at 09:13 PM

So what we would do is have eratio for each system, and then do nth stats on the good eratio and compare with the bad eratio? THat would show if the concept works or not. Im open to other ideas?

Yes that would be an interesting test.

could also be used for looking at exits, as the article states,

(Additionally, if E-Ratio falls off a cliff at bar 6… then it probably does not make sense to hold for 15 bars!)

admin - 17-10-2018 at 09:16 PM

So what we would do is have eratio for each system, and then do nth stats on the good eratio and compare with the bad eratio? THat would show if the concept works or not. Im open to other ideas?

Personally I think that exit time is going to be really simple. It wont be linked so much as to exit after x bars, but exit at a certain time. (Day trading)
Swing trading could be a different matter.

waldocktrades - 18-10-2018 at 07:23 AM

Right now, I've got it set to look ahead two days ahead and as far back as I'd like. I average the bar by bar net change by timeframe and lookback in the same subgraph to provide an expected shape to the day's trading. This creates a forward-looking indicator based on intraday market behavior.
I think it could be very interesting to see what GSB could do with it.
If you're interested, I can email you more.

admin - 18-10-2018 at 07:55 AM

Quote: Originally posted by waldocktrades

Right now, I've got it set to look ahead two days ahead and as far back as I'd like. I average the bar by bar net change by timeframe and lookback in the same subgraph to provide an expected shape to the day's trading. This creates a forward-looking indicator based on intraday market behavior.
I think it could be very interesting to see what GSB could do with it.
If you're interested, I can email you more.

Most happy to see it, please send

admin - 18-10-2018 at 08:54 PM

.40 under testing. Has verification score
Soon we will have market verification/degradation score per system.
So lets say you build 5000 CL systems on 29,30,31 min bars.
You can then verify on 20,24...36,40 min bars. (12 different bar intervals)
You would expect the 20,24...36,40 min bars to give a bit worse performance than the 29,30,31.
Then those systems that get verification score of say 10 up, or even 12.
You can then choose the system where the verification degradation is the highest.
Shown here the verified bars had higher performance than the 29,30,31 that the system was build on.
The out of sample (alternate nth day) also had higher results than in sample.
I would say this would be very unusual.
800 systems were tested to get this result.

BUG in versions since .35 on-wards. (but much less bugs in these newer versions, esp TS<>GSB)
Use built in setexitonclose if true, gives exception error on some systems, esp when verifying them.

Relates to 1/2 trading days.
fix setexitonclose =false

admin - 19-10-2018 at 04:56 PM

49.44 out. Has possible bug fix for wf issue, which relates to 1/2 trading days bug.
Now internal and external market on close should be working again.
Option to turn certain curves off and on.
under tools, reset price data cache. This reloads price data if you want to update your csv data
Build removed. Has critical bug

admin - 20-10-2018 at 12:09 AM

49.45 out. See top of this post.

admin - 5-11-2018 at 04:21 PM

GSB 50.06 Nov 11 2018
Has genetically chosen stop. Static, trailing and atr.
Build not well tested.
Issues.
1)New stop settings dont save in config.
2) Dont use with stop loss of zero. Use say $4000. Reason is stop will be min (new stop method,stop size under exits.)
This build is NOT well tested. Main reason to get it is due to new stops. Stops will be expanding over time, but this is the first release.
3) I suspect either Microsoft or avg is deleting the exe file on one of my machines. (false positive)
stops.png - 65kB

Bruce - 6-11-2018 at 04:31 AM

GSB 50.06 Nov 11 2018
Has genetically chosen stop. Static, trailing and atr.
Build not well tested.
Issues.
1)New stop settings dont save in config.
2) Dont use with stop loss of zero. Use say $4000. Reason is stop will be min (new stop method,stop size under exits.)
This build is NOT well tested. Main reason to get it is due to new stops. Stops will be expanding over time, but this is the first release.
3) I suspect either Microsoft or avg is deleting the exe file on one of my machines. (false positive)
stops.png - 65kB

I've excluded the GSB directory from Microsoft Defender to stop the deleting of the exe file. Seems to have stopped it for now.

rws - 6-11-2018 at 01:49 PM

Thanks for adding stops!

admin - 7-11-2018 at 12:51 AM

Quote: Originally posted by rws

Thanks for adding stops!

Let me know if there is much improvement.
I dont expect atr and trailing stops to work well on crude oil day trading,
more for the longer term trades.
Im not doing any of that for some time, but interested in the results.

admin - 19-11-2018 at 04:12 AM

50.14 19 nov 2018
Has MOC exit if openpositionprofit <$0, exit if daily close is lower than last close.
Both for overnight systems.
I have not tested this build at all, and the ts code has not been tested either
Option to cancel verification.
In next build, auto verify systems when they are built.

cotila1 - 19-11-2018 at 12:21 PM

So if I want to build only multiday systems, always overnight, then I need to set both MarketOnDay=False and the 2 new exits set to false as well??

The auto-verify of built systems functionality is something I cant wait to see it :-)

50.14 19 nov 2018
Has MOC exit if openpositionprofit <$0, exit if daily close is lower than last close.
Both for overnight systems.
I have not tested this build at all, and the ts code has not been tested either
Option to cancel verification.
In next build, auto verify systems when they are built.

admin - 20-11-2018 at 12:54 AM

Quote: Originally posted by cotila1

So if I want to build only multiday systems, always overnight, then I need to set both MarketOnDay=False and the 2 new exits set to false as well??

The auto-verify of built systems functionality is something I cant wait to see it :-)

If overnight you must have marketonclose false.
Its optional to have exit that if openpostionprofit<0, it takes the moc exit.
and if the daily close closes down. There is a methadolgyt bug in the later.
It should be if entriestoday(date)= 0 and daily close down then exit
I found out today that auto verification takes all the cpu manager resource, and causes lots of problems. ie you cant stop the manager, and the manager drops workers. Sorry but the last build was un-tested when I uploaded.
There also is a small ts compile issue with duplicate exit names. SO beset ignore the last build sorry.

cotila1 - 20-11-2018 at 04:09 AM

Quote: Originally posted by cotila1

Got it! thanks

admin - 21-11-2018 at 12:44 AM

IM going to remove 50.15 build as I think its too problematic. Next build will have verification on the workers, not just the manager.
When it was on the manager, the GUI became unusable slow to respond, and ram usage can really blow out. The ram useage itself is not a bug, just related to how many versifications are done in one time.

admin - 21-11-2018 at 07:39 PM

50.18 Nov 22 2018
verify systems in the workers seemed to go well, at quick test.
you must update all dlls in this version. There should be no config in the zip file that over writes your own config.
fix for busted WF.
New exits, ie moc if openpostionprofit<0 not fixed yet

cyrus68 - 22-11-2018 at 01:25 AM

The new exit methods are fine. However, what is also needed is a maximum holding-period exit. In other words, an exit is executed X bars after entry. This can be applied to both daily and intraday bars.

admin - 22-11-2018 at 04:43 AM

The new exit methods are fine. However, what is also needed is a maximum holding-period exit. In other words, an exit is executed X bars after entry. This can be applied to both daily and intraday bars.

We have that already. Exit minutes.

Gregorian - 22-11-2018 at 10:26 AM

We have that already. Exit minutes.

Those of us who trade with Kase and Renko bars would appreciate a number of bars exit, as cyrus68 suggests.

admin - 22-11-2018 at 03:51 PM

Quote: Originally posted by Gregorian

We have that already. Exit minutes.

Those of us who trade with Kase and Renko bars would appreciate a number of bars exit, as cyrus68 suggests.

So you mean number of bars, NOT number of minutes?

admin - 22-11-2018 at 06:03 PM

There is a bug in 50.18
On older manager, when I had verify auto on in the manager (not the current mode of verify in the worker), the manager became very unresponsive and lost systems from the worker. As a backup on 50.18 each 20 min we pulled in the systems again from the worker. This made a bug where we can pull in stale systems that are still in the worker from another manager.
I will get this fixed asap. 50.18 overall is a good build but has a few issues still to sort out.

Gregorian - 22-11-2018 at 06:33 PM

So you mean number of bars, NOT number of minutes?

Yes, number of bars exit would be great.

admin - 22-11-2018 at 06:36 PM

Quote: Originally posted by Gregorian