| Pages:
1
..
13
14
15
16
17
..
98 |
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Quote: Originally posted by eatrade  | I just installed the trial file GSB 47.44 build. July 24th 2018 and am surprise,d that the duration ist set to only 4 days left from the beginning. I
thought that the trial period is 14 days. Is this a bug or indeed only 4 days to try the software?
Thank you very much in advance, |
This is a bug, and I think its fixed in the next version. However once you install your current version, the new version wont fix the issue.
I will email you a new file to fix this. Apologies for a bug.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
I haven't tried to test systems on GSB for a good while. But seeing the possibility to test via verification price data, I decided to give it a try. I
already had a successful NG model that produced some very attractive systems. NG as data1, plus data2, 3 and 4. It ran smoothly on 2 computers - the
only glitch was that the manager did not aggregate all unique systems correctly.
I ran the model with the SAME SETTINGS as previously, except that verification data was included (NG as data1, plus data2, 3 and 4 - at a different
frequency).
Problem #1. GSB produced no systems, nothing. I excluded the verification data. Still nothing.
Problem #2. on the 2nd computer running 6 workers, only one of the workers had copied the correct app and opt settings. The rest were running on opt
and app settings from a previous run.
This was on build 47.49. Now, my problem is to figure out what was the last build that worked properly. I am not interested in running the test data
and settings that come with GSB, but my own data and settings.
As I haven't run anything on GSB for a good many weeks, I will have to try various exe files and see which one works with the fewest bugs. Another
issue is whether the old exe files will be compatible with the other files in the installed 47.49 directory.
All in all, an unfortunate state of affairs.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | I haven't tried to test systems on GSB for a good while. But seeing the possibility to test via verification price data, I decided to give it a try. I
already had a successful NG model that produced some very attractive systems. NG as data1, plus data2, 3 and 4. It ran smoothly on 2 computers - the
only glitch was that the manager did not aggregate all unique systems correctly.
I ran the model with the SAME SETTINGS as previously, except that verification data was included (NG as data1, plus data2, 3 and 4 - at a different
frequency).
Problem #1. GSB produced no systems, nothing. I excluded the verification data. Still nothing.
Problem #2. on the 2nd computer running 6 workers, only one of the workers had copied the correct app and opt settings. The rest were running on opt
and app settings from a previous run.
This was on build 47.49. Now, my problem is to figure out what was the last build that worked properly. I am not interested in running the test data
and settings that come with GSB, but my own data and settings.
As I haven't run anything on GSB for a good many weeks, I will have to try various exe files and see which one works with the fewest bugs. Another
issue is whether the old exe files will be compatible with the other files in the installed 47.49 directory.
All in all, an unfortunate state of affairs. |
Just do a support upload, and I will look at it. Are all the workers the same version? old managers typically can drive newer build workers, but not
the other way around. So far no other user complaints, but you can always go back to the last version you were happy with.
|
|
|
eatrade
Junior Member
Posts: 2
Registered: 23-7-2018
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by eatrade | I just installed the trial file GSB 47.44 build. July 24th 2018 and am surprise,d that the duration ist set to only 4 days left from the beginning. I
thought that the trial period is 14 days. Is this a bug or indeed only 4 days to try the software?
Thank you very much in advance, |
This is a bug, and I think its fixed in the next version. However once you install your current version, the new version wont fix the issue.
I will email you a new file to fix this. Apologies for a bug. |
The trial period has not changed, it is no longer available today, I also installed 47.49a...
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by eatrade | | Quote: Originally posted by admin | | Quote: Originally posted by eatrade | I just installed the trial file GSB 47.44 build. July 24th 2018 and am surprise,d that the duration ist set to only 4 days left from the beginning. I
thought that the trial period is 14 days. Is this a bug or indeed only 4 days to try the software?
Thank you very much in advance, |
This is a bug, and I think its fixed in the next version. However once you install your current version, the new version wont fix the issue.
I will email you a new file to fix this. Apologies for a bug. |
The trial period has not changed, it is no longer available today, I also installed 47.49a... |
While I hope that 44.49a has fixed the expiry bug, if you have installed 49.44 first, then 49.49 will NOT fix the expiry.
please send me your teamviewer.com details and I will fix this asap. Sorry ofr inconvenience caused. I cant test this issue myself as I have no
computers at all that have never had GSB run on them. info@trademaid.info
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
A month ago I could generate systems on GSB, with very few glitches. Coming back to GSB after a month, I encounter one problem after another.
Previously, I could load saved systems and re-examine them. Now, it is impossible and I get multiple error messages. Also, as noted before, the same
settings on the same data-sets, that generated good results, now produce nothing. And there is a problem with workers being unable to read the
manager's settings that has nothing to do with different versions.
I have used GSB for over a year. So I'm not a novice. Right now, for me, this software has lost functionality. I can't do basic things that were
possible previously. The most important thing for this sort of software is reliability and consistency. If I follow consistent rules in using the
software, then I should see consistent outcomes. These are serious issues that need to be addressed.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | A month ago I could generate systems on GSB, with very few glitches. Coming back to GSB after a month, I encounter one problem after another.
Previously, I could load saved systems and re-examine them. Now, it is impossible and I get multiple error messages. Also, as noted before, the same
settings on the same data-sets, that generated good results, now produce nothing. And there is a problem with workers being unable to read the
manager's settings that has nothing to do with different versions.
I have used GSB for over a year. So I'm not a novice. Right now, for me, this software has lost functionality. I can't do basic things that were
possible previously. The most important thing for this sort of software is reliability and consistency. If I follow consistent rules in using the
software, then I should see consistent outcomes. These are serious issues that need to be addressed. |
Sorry you are frustrated. Until you do a support upload and or do teamviewer, I cant help you. Im using 53 workers right now, and its going fine for
me. I loaded system a system from nov 2017 into 48.00 and it worked fine.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
I would like to see solutions to some of the problems, whichever method works. I am not using your experimental data or settings.
Also, for the first time, I am seeing an error message regarding contracts. What are loaded contracts? My understanding is that you need to set the
contract specs in the contract list under 'Tools".
The file name is NG.15.Minute.900.1430.NG_datexxx, which refers to the optimal exchange session time. My computer is on local time= central time US.
So the appropriate MOC is set at 1330.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | I would like to see solutions to some of the problems, whichever method works. I am not using your experimental data or settings.
Also, for the first time, I am seeing an error message regarding contracts. What are loaded contracts? My understanding is that you need to set the
contract specs in the contract list under 'Tools".
The file name is NG.15.Minute.900.1430.NG_datexxx, which refers to the optimal exchange session time. My computer is on local time= central time US.
So the appropriate MOC is set at 1330. |
That error shows your local contracts.txt file contained 1430, but your data file 1330. GSB wants to update this to 1330 (correct time)
GSB has hard coded times and contract data that will be automatically over written if you have your own contracts.txt file.
For other forum users, Cyrus's GSB did not appear to make systems. The metrics in training were loosened off, and the full period performance metrics
loosened too. If you have nth set to 1, np and fitness should 1/2 as we have 1/2 the amount of trading days. However pearsons will also drop and the
system is less likely to be as linear.
Also for NG we changed secondary filter to only user closed/closed, not the other 36 secondary filters. Normally if you use GA S.F. it will choose
closed(minus or divide) regardless, and if it doesnt, the systems choose are not as good as CLoseD
For NG close/CloseD was the best S.F.
You can check whats going on in workers or stand alone by right clicking the top diag or latest diag tab and clicking refresh.
Takes a few minutes to produce systems.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
Looking past the glitches, the new verification feature is excellent. New to me, because I hadn't used GSB for a good number of weeks.
I got another pop-up notification on contracts that makes little sense. I loaded the manager to test ES systems. First, I saved the new app settings.
Then, I tried to save the opt settings and the notification popped up. What has NG got to do with it?
The correct specs for both ES and NG have been saved in the contracts list, a year ago. GSB is supposed to read the contract specs from that list.
This brings me to the conventional nomenclature for naming data files (txt) in GSB. The begin and end session times are based on the optimal custom
settings in TS, based on the contract's exchange time. Your computer's time-zone setting will determine the actual begin and end session times, during
which trading will take place. So, if your txt file is called NG.15.Minute.900.1430.NG_datexxx, the custom session refers to US Eastern time. And, if
your computer is on US Central time, this means 800 to 1330. The data file will reflect the session times for Chicago, not NY.
So, where is the error? In the file name or somewhere else?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | Looking past the glitches, the new verification feature is excellent. New to me, because I hadn't used GSB for a good number of weeks.
I got another pop-up notification on contracts that makes little sense. I loaded the manager to test ES systems. First, I saved the new app settings.
Then, I tried to save the opt settings and the notification popped up. What has NG got to do with it?
The correct specs for both ES and NG have been saved in the contracts list, a year ago. GSB is supposed to read the contract specs from that list.
This brings me to the conventional nomenclature for naming data files (txt) in GSB. The begin and end session times are based on the optimal custom
settings in TS, based on the contract's exchange time. Your computer's time-zone setting will determine the actual begin and end session times, during
which trading will take place. So, if your txt file is called NG.15.Minute.900.1430.NG_datexxx, the custom session refers to US Eastern time. And, if
your computer is on US Central time, this means 800 to 1330. The data file will reflect the session times for Chicago, not NY.
So, where is the error? In the file name or somewhere else? |
The GSB format is symbol.interval.type.anythinguLike.txt
What im doing myself for symbols that can have mutiple close times is
symbolMOC.minute.CustomSessionTimeStart_to_CustomSessionTimeEnd_timezone_startDate_endDate.txt
CL1330.30.Minute.900_TO_1430_centralUSAtime_200609_20180609
You then need to copy the CL contract under contracts from CL to CL1330 and set the Moc time to 1330.
What moc time do you have in C:\GSB\GSB (Managers)\data\contracts.txt
Bottom line is you should click yes to update.
All this session time stuff is confusing for most users (even me) as TS Custom session time refers to exchange time, but the bars are written local
time.
Hope this reply helps. Thanks for the comments on how well verification data works.
|
|
|
ProbTrader
Junior Member
Posts: 16
Registered: 3-7-2018
Member Is Offline
Mood: No Mood
|
|
Hi all
Has anybody put thinking/analysis into what the "best" fitness functions for OOS/WF might be, and incorporated the fact that this may differ between
instruments/asset classes? Its easy to enough to choose a fitness function that produces a good looking IS equity curve (eg smooth with low drawdowns
and high CAGR etc) but I have found that the preferred fitness functions for IS are not necessarily and generally the same as the preferred for OOS.
In other words, fitness function X may produce the best IS curves, but fitness function Y may actually hold better OOS. Thanks.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
My file names and contract specs have been the same since last year. My computer is on central time and the MOC times reflect this fact. I re-checked
them and they are correct. I reiterated the relationship between exchange time, TS custom session time and the computer's local time to avoid any
confusion and pinpoint possible changes in GSB that I'm not aware of.
The error notification refers to NG when I am setting up a system for ES. Bizarre indeed. In any case, if this is a harmless new bug, I can ignore it.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by ProbTrader | Hi all
Has anybody put thinking/analysis into what the "best" fitness functions for OOS/WF might be, and incorporated the fact that this may differ between
instruments/asset classes? Its easy to enough to choose a fitness function that produces a good looking IS equity curve (eg smooth with low drawdowns
and high CAGR etc) but I have found that the preferred fitness functions for IS are not necessarily and generally the same as the preferred for OOS.
In other words, fitness function X may produce the best IS curves, but fitness function Y may actually hold better OOS. Thanks.
|
My post today on CL implied that pearsons was much better than NP*AT
http://trademaid.info/forum/viewthread.php?tid=117
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by ProbTrader | Hi all
Has anybody put thinking/analysis into what the "best" fitness functions for OOS/WF might be, and incorporated the fact that this may differ between
instruments/asset classes? Its easy to enough to choose a fitness function that produces a good looking IS equity curve (eg smooth with low drawdowns
and high CAGR etc) but I have found that the preferred fitness functions for IS are not necessarily and generally the same as the preferred for OOS.
In other words, fitness function X may produce the best IS curves, but fitness function Y may actually hold better OOS. Thanks.
|
My post today on CL implied that pearsons was much better than NP*AT
http://trademaid.info/forum/viewthread.php?tid=117 |
Was the fitness criteria set as NP*Avg trade*Pearson or simply Pearson?
I would like to try NP*Avg trade*Sharpe, but haven't yet.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | | Quote: Originally posted by admin | | Quote: Originally posted by ProbTrader | Hi all
Has anybody put thinking/analysis into what the "best" fitness functions for OOS/WF might be, and incorporated the fact that this may differ between
instruments/asset classes? Its easy to enough to choose a fitness function that produces a good looking IS equity curve (eg smooth with low drawdowns
and high CAGR etc) but I have found that the preferred fitness functions for IS are not necessarily and generally the same as the preferred for OOS.
In other words, fitness function X may produce the best IS curves, but fitness function Y may actually hold better OOS. Thanks.
|
My post today on CL implied that pearsons was much better than NP*AT
http://trademaid.info/forum/viewthread.php?tid=117 |
Was the fitness criteria set as NP*Avg trade*Pearson or simply Pearson?
I would like to try NP*Avg trade*Sharpe, but haven't yet. |
fitness was np*at only
|
|
|
ProbTrader
Junior Member
Posts: 16
Registered: 3-7-2018
Member Is Offline
Mood: No Mood
|
|
I think the choice of fitness function is one of the core assumptions of any system you build. If you're wrong here, every next step of the process
will rest on a bad assumption. And the research I have done suggests that the optimal fitness function can vary greatly from case to case (obviously
beware of over-fitting etc), the same way that any other system input can and do.
Two such fitness criteria I think GSB would benefit from having as an output in the "lower output window"/GUI are Sharpe and the t-test.
|
|
|
Petzy
Junior Member
Posts: 73
Registered: 24-10-2017
Location: Sweden
Member Is Offline
Mood: No Mood
|
|
I have tried pearson * NetProfitOverDrawdown. It seems to work quite ok. But I have not yet finnished testing
|
|
|
rws
Member
Posts: 114
Registered: 12-6-2017
Member Is Offline
Mood: No Mood
|
|
You can have great fitness if you allow very few trades. The fitness and the number of trades are both important. Too few trades have no statistical
meaning and are easier to curve fit.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
rws is correct. There is also another reason why more trades are better than fewer. If your system deteriorates in live trading, you can pick this up
sooner if the system executes many trades. This allows you to act in stopping the losses.
As for the Sharpe Ratio, in this context, it measures return per unit of volatility, not risk. Drawdown is a better measure of risk. If the division
operator was available in GSB for setting fitness criteria, we could construct NP/(Avg of top 5 drawdowns).
I am not in favour of complex fitness functions. The criteria that are included should make sense. For example, %profitable and PF both measure the
success rate, and it is redundant to include both.
A while back, I was searching for a GSB video and came across one by TSL, which I don't own. Here is a list of their fitness function criteria. They
have pretty much stacked the shelf with everything. Note that #30 is the Sharpe Ratio + Total Trades. A number of the others use the addition
operator.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
I just switched from 47.49 to 48.16. When I want to insert a dataset in the relevant slot and click on "save", I am immediately presented with a huge
list of contracts and indices, telling me that they are invalid price file contracts. Never mind that this is irrelevant to the dataset I am trying to
insert.
Essentially it is reading all contracts and indices in the price data list and checking to see if it is in the contracts list. I can't ignore the
message because it won't allow me to insert the selected dataset. Actually, I am not interested in verifying on other markets. Also, I use a lot of
indices and contracts as secondary data. Entering them on the contracts list is, for me, unnecessary and a pain in the neck. How is it possible to
disable or get past this nasty message? Right now, I have returned to 47.49.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | I just switched from 47.49 to 48.16. When I want to insert a dataset in the relevant slot and click on "save", I am immediately presented with a huge
list of contracts and indices, telling me that they are invalid price file contracts. Never mind that this is irrelevant to the dataset I am trying to
insert.
Essentially it is reading all contracts and indices in the price data list and checking to see if it is in the contracts list. I can't ignore the
message because it won't allow me to insert the selected dataset. Actually, I am not interested in verifying on other markets. Also, I use a lot of
indices and contracts as secondary data. Entering them on the contracts list is, for me, unnecessary and a pain in the neck. How is it possible to
disable or get past this nasty message? Right now, I have returned to 47.49. |
This was in red on the release notes. As we have closedbpv, you must add all secondary data into your contracts list.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
Quelle horreur.
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
Has anybody been able to generate walk-forward results in build 48.16?
Let’s take the CL example used in the video, for the sake of familiarity. You load the 29 30 and 31 min datasets, set “Opt price data” to False and
generate systems.
Next, you want to run WF on a given system. So, you load the 30 min dataset in “WF price data”. Then, you run the WF. Except that, on completion, you
don’t see the OOS and Current curves, no reports and no script.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | Has anybody been able to generate walk-forward results in build 48.16?
Let’s take the CL example used in the video, for the sake of familiarity. You load the 29 30 and 31 min datasets, set “Opt price data” to False and
generate systems.
Next, you want to run WF on a given system. So, you load the 30 min dataset in “WF price data”. Then, you run the WF. Except that, on completion, you
don’t see the OOS and Current curves, no reports and no script.
|
You should be on 48.24 build. Uploaded in the beta section.
WF is not perfect but getting better. The only current bug I am aware of is often wf to cloud will be on nth=1 not nth = all. This is always the case
for me, but is linked to something in config and will work fine for some users. Wf to local didnt have this bug
|
|
|
| Pages:
1
..
13
14
15
16
17
..
98 |
![[*]](images/davis/default_icon.gif){kind=icon}
