| Pages:
1
2
3
4 |
cotila1
Junior Member
Posts: 78
Registered: 8-5-2017
Member Is Offline
Mood: No Mood
|
|
I like this new CloseD (CLoseDBPV) . That means that if I use the close-closeD(1) as SF for ES and I want to verify on nq, then i should not have big
difference/issue by using this new CLoseDBPV?
So the old ''CloseLessPrevCLoseD'' will have new name in SF dropdown menu?
Quote: Originally posted by admin  | A new release will be out for trial users by the end of the month. Old code will have expired too.
We now have CLoseDBPV (Big point value) which makes closeD-Close redundant. Much better for when we you compare multiple markets that have very
different point values. Lots of small tweaks and features have been added. Much better use of ram when nth / dates tests are used.
There now is a job que on the right side gui for NTH, dates and market verification. (49.09 onwards)
A help file. (not complete) has been added under help. (top menu)
|
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cotila1 | I like this new CloseD (CLoseDBPV) . That means that if I use the close-closeD(1) as SF for ES and I want to verify on nq, then i should not have big
difference/issue by using this new CLoseDBPV?
So the old ''CloseLessPrevCLoseD'' will have new name in SF dropdown menu?
| Quote: Originally posted by admin | A new release will be out for trial users by the end of the month. Old code will have expired too.
We now have CLoseDBPV (Big point value) which makes closeD-Close redundant. Much better for when we you compare multiple markets that have very
different point values. Lots of small tweaks and features have been added. Much better use of ram when nth / dates tests are used.
There now is a job que on the right side gui for NTH, dates and market verification. (49.09 onwards)
A help file. (not complete) has been added under help. (top menu)
|
|
correct on all points but dropdown box will have closed and closedBPV
|
|
|
cotila1
Junior Member
Posts: 78
Registered: 8-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by cotila1 | I like this new CloseD (CLoseDBPV) . That means that if I use the close-closeD(1) as SF for ES and I want to verify on nq, then i should not have big
difference/issue by using this new CLoseDBPV?
So the old ''CloseLessPrevCLoseD'' will have new name in SF dropdown menu?
| Quote: Originally posted by admin | A new release will be out for trial users by the end of the month. Old code will have expired too.
We now have CLoseDBPV (Big point value) which makes closeD-Close redundant. Much better for when we you compare multiple markets that have very
different point values. Lots of small tweaks and features have been added. Much better use of ram when nth / dates tests are used.
There now is a job que on the right side gui for NTH, dates and market verification. (49.09 onwards)
A help file. (not complete) has been added under help. (top menu)
|
|
correct on all points but dropdown box will have closed and closedBPV |
So If i want to use close-closeD(1) as SF for ES building and I want to verify on nq and avoid the BPV issue between es and nq, should I choose closed
or closedBPV for building?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cotila1 | | Quote: Originally posted by admin | | Quote: Originally posted by cotila1 | I like this new CloseD (CLoseDBPV) . That means that if I use the close-closeD(1) as SF for ES and I want to verify on nq, then i should not have big
difference/issue by using this new CLoseDBPV?
So the old ''CloseLessPrevCLoseD'' will have new name in SF dropdown menu?
| Quote: Originally posted by admin | A new release will be out for trial users by the end of the month. Old code will have expired too.
We now have CLoseDBPV (Big point value) which makes closeD-Close redundant. Much better for when we you compare multiple markets that have very
different point values. Lots of small tweaks and features have been added. Much better use of ram when nth / dates tests are used.
There now is a job que on the right side gui for NTH, dates and market verification. (49.09 onwards)
A help file. (not complete) has been added under help. (top menu)
|
|
correct on all points but dropdown box will have closed and closedBPV |
So If i want to use close-closeD(1) as SF for ES building and I want to verify on nq and avoid the BPV issue between es and nq, should I choose closed
or closedBPV for building? |
Use bpv, there is no need to use close-CloseD any more. Beta 49.09 uploaded in private forum shortly
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
49.12 has comments in data table. They are only for the user, not gsb.
There always is lots of confusion over session times.
Your TS should run local time. This is because if you have a data 2 with different time zone to data1, TS&MC wont allow it.
The session time on your chart is exchange time, but the exported data is in your local time.
Moc time should = local time last bar of day, not exchange time.
Note also cash indices have changed point value to what is on the futures contract. This is needed for closeDBPV (big point value) to work best
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
In the beta section (GSB paid members) 49.20 was released today. Lots of long awaited features.
One of the best things in GSB is the ability to build 1000's of systems. Then tweak things to see what improves the out of sample results. We were
using every second day as un seen to get a out of sample picture. Then we look at how much out of sample degrades compared to in sample. Well now we
can choose say 3 days to trade, 1 day to not trade. This means greater in sample period - so we get more trades to form a more valid opinion. We can
then get just as many trades out of sample by twice the amount of trades. (Before we traded 1 day, then 1 day out of sample) So now the statistics are
normalized to take into account that you have less trading days.
We can also have unlimited custom dates for training, test and validation.
There is now a job que to submit changes in the dates or nth so changes in dates and nth can be done in one pass, not two.
|
|
|
kelsotrader
Junior Member
Posts: 29
Registered: 16-2-2018
Location: Tapanui - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | 49.12 has comments in data table. They are only for the user, not gsb.
There always is lots of confusion over session times.
Your TS should run local time. This is because if you have a data 2 with different time zone to data1, TS&MC wont allow it.
The session time on your chart is exchange time, but the exported data is in your local time.
Moc time should = local time last bar of day, not exchange time.
Note also cash indices have changed point value to what is on the futures contract. This is needed for closeDBPV (big point value) to work best
|
Excellent Explanation: I always add the exchange session time onto the file name when exported so that the GSB code will record the time in its code
so as to keep a permanent record on the strategy.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Soon we will have market 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.
 
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
We can now turn off and on specific groups of curves, and the legend. This stops the graphs being too cluttered.
In build 49.44 onwards
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
I agree entirely. I think the information on the graph can be revised to reduce clutter and be more relevant. Instead of all the line items showing
the metrics for each verification dataset (whether it can be turned on or off), we could have a single one showing the average metrics for all the
verification datasets. This will allow us to make a direct comparison with the average result of the datasets that were used in the build. Is this
what the VSS score is showing?
Having reduced the clutter, it is important to include line items for the metrics of EACH of the datasets that were used in the build – in this case,
29 30 31 min. I would like to see both the IS and OOS results. This is essential information when we want to run WF. I don’t automatically run WF on
the central bar size and, even if I do, I want to see its metrics – not just the average IS and OOS for all of them.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | I agree entirely. I think the information on the graph can be revised to reduce clutter and be more relevant. Instead of all the line items showing
the metrics for each verification dataset (whether it can be turned on or off), we could have a single one showing the average metrics for all the
verification datasets. This will allow us to make a direct comparison with the average result of the datasets that were used in the build. Is this
what the VSS score is showing?
Having reduced the clutter, it is important to include line items for the metrics of EACH of the datasets that were used in the build – in this case,
29 30 31 min. I would like to see both the IS and OOS results. This is essential information when we want to run WF. I don’t automatically run WF on
the central bar size and, even if I do, I want to see its metrics – not just the average IS and OOS for all of them.
|
VSS is showing the metric for the individual time frame . market -not the average. I will look into "see both the IS and OOS results". This will be
even more info in the GUI.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Additional work on exits has started about 2 weeks ago. Hence the lack of other updates the last few weeks.
I'm nearly finished the next video on CL, with significant refinement of methodology.
The next release will appear in the beta forum for paid users.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
In the beta forum, we now have trailing stops, atr stops, exits after x bars (we has exits after x minutes some time ago)
Additional exits are being debugged now.
The new video is out. https://www.youtube.com/watch?v=HDeJpONE090
verification of systems done in the workers, instead of only in GSB managers.
Soon we should have counter trend oscillators which should result in counter trend systems
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
GSB counter trend oscillator released in the private forum today.
|
|
|
JasonT
Junior Member
Posts: 61
Registered: 6-6-2018
Member Is Offline
Mood: No Mood
|
|
Very Impressed
Hi Peter,
I just wanted to say how impressed I am with the 50.25 beta version of GSB. I'm finding it much easier to use than the 49.xx series, especially the
way you've incorporated the IS/OOS degradation statistics - fantastic.
I've gone through the last 3 you tube videos a few times over the weekend and have got a much better idea of how to perform degradation testing. It
took me a while to piece it all together. Now that I understand what you are doing and why, it is straight forward. Good work.
Videos here:
a) Genetic System Builder Market testing / validation 22 May 2018
b) GSB Market Validation video2 2 Jul 2018
c) Genetic system builder, new features to get better out of sample results 1 Nov 2018
|
|
|
JasonT
Junior Member
Posts: 61
Registered: 6-6-2018
Member Is Offline
Mood: No Mood
|
|
I'm also a fan of the User Tasks section in the resource monitor panel on the right. It's great to see progress against a change of Nth day mode or
Verification calculation.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Short term updates.
Coming soon.
Subtle improvements to the allocation of GSB cloud.
Export of all the metrics of say just 30 minute bars, even if we build on 29,30,31
various bug fixes
Not to soon we start on new/additional secondary filters.
GSBramv2 is a bit over a week away. This will allocate your idle cpu to the general cloud resulting in more cloud for all.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Resource manager (RM) is running well, and there now is decent cloud power for all. RM is contributing unused CPU power from GSB users, and I have 4
high end machines that often contribute to the cloud.
Updates to RM are planned. Ie run on computer startup. Start x workers on startup, rather than starting just 1.
Additional price, time filters are being built right now. We will start with a small library just to test that it all works, then increase the
features.
This hopefully will make a much more powerful GSB. BEST case the first beta build will be ready at the end of the week.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
This is what the next build of GSB looks like. Likely these new groups in cyan will be combined into one function each.
This is not a exhaustive list, just a few added for proof of concept.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
GSB 2.0
I have been asked about GSB 2.0
There are a number of things that need to be revamped, and a lot of that is to make user errors less probable.
Richard Bransons space shuttle crashed, just because someone pressed a button at the wrong time. GSB issues are much more minor, but this section
needs a revamp.
The first version of this is basically going to have the secondary filters reduced to closeD variants,
and the normalized and non normalized are all going to be in the same set of filters.
Currently you have non normalized CLoseD.. or GA (which are all normalized) with closeD and 37 or so indicators.
(Normalization is where we adjust the last 100 bars of data to be in a range of -100 to +100)
The number one issue users make is wrong secondary filter (SF). It is critical to get this right.
Some users also add all 38 SF. This results in much slower build of systems, and the systems that dont chose a Closed SF, work much poorer.
So SF are going to be soft coded to the contract used. So if a user does nothing, the setting will be correct.
So this section is simplified, and secondary filters could ALL be chosen genetically. Currently this cant be done.
We will keep compatibility with old systems in new managers of GSB 2.0, but we will loose optimizer setting compatibility.
app settings will be the same.
Opt settings needs the return of secondary filter closed parameters, but a different set of values if closed-close vs closed/close.
These sorts of things bit too many users, which is why we hid them and made them auto detect. Other uses wanted control of this variable.
Likely we also need to have the indicator weights linked to what operands we use.
ie if operand * is used, we use weight -1 to 1 step 2 and the others set to 1
if addition is used, we might use -100 to 100 step 10 etc
For new users, the over simplified GSB system format is
result=indicator1*weight1{operand = */+}indicator2*weight2{operand = */+}indicator3*weight3{operand = */+}
secondary filter =closeD(1)-close
if result >0 and secondaryFilter=true then buy..
Im also hoping to add the ability for gsb to make the binary files that EWFO and ts wfo.exe makes.
My wild guess is GSB is going to be 10 to 100 times faster to make these.
EWFO has a number of features that gsb wf doesnt have. Cluster analysts, semi automatic fitness detection.
Wf efficiency (where out of sample has to be reasonable relative to in sample)
These are not essential features, but a number of users have expressed interest in this.
The big area of GSB growth will be in tertiary filters. Tertiary means 3, so the GSB over simplied system format will have added
if result >0 and secondaryFilter=true and Tertiaryfilter=true
then buy..
There is massive scope in Tertiary filters, and we will do a few things, test for bugs, then continue to expand
|
|
|
coccigelus
Junior Member
Posts: 73
Registered: 11-7-2018
Member Is Offline
Mood: No Mood
|
|
Very interesting the new many directions GSB is taking. I think the most interesting feature will be the implementing of the price pattern and could
really open a lot of new doors and non correlated systems including to avoid some critics things involved with oscillators. That area should be easily
expanded later increasing the number of candles analyzed.
Regarding the tertiary filters I am skeptical for a couples of reasons.
Kudos to the work You are doing on the opt/fitness/normalization criteria. That is really important and really needed including increase flexibility
for the users.
THX for feed back.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Whats almost finished is the ability to walk forward gsb systems in GSB, but send the output to EWFO
https://trademaid.info/hugo/ewfo.html
GSB WF is fast, but EWFO WF is much more powerful. This gives you the best of both worlds.
|
|
|
cotila1
Junior Member
Posts: 78
Registered: 8-5-2017
Member Is Offline
Mood: No Mood
|
|
that's a very clever feature. thank you
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
This is whats new in build 52.32
Macros instant cancel implemented.
Macros logging (minimal to text file) implemented.
The sorting and top added to favorites macro.
Sound macro implemented.
Email macro implemented.
FavoritesReset macro and right-click button implemented.
StatsReset macro and right-click button implemented.
Stats G-H implemented.
WF Params. (separate from Param.) implemented.
Subtract operator added.
Find (system by ID) button implements.
Improved GUI responsiveness, CPU and RAM on managers (by disabling backtest on new systems download).
Allowed Nth invert (and similar tasks) to run multi-threaded while optimization is active.
Multiple fixes in cloud WFs.
Selected Rows/Cols/Cells implemented.
Minor GUI updates and fixes.
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
Hi Peter,
I have not had a chance to update and test just yet, but by "WF Params. (separate from Param.) implemented", are you referring to the ability to save
and restore WF params when saving systems that have been walked forward, or something else?
|
|
|
| Pages:
1
2
3
4 |
![[*]](images/davis/default_icon.gif){kind=icon}
