MULTICHARTS 12.0 RELEASE 4

Questions about MultiCharts and user contributed studies.
User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 28 Dec 2018

MultiCharts 12.0 Release 4 builds 17488 (x32) / 17490 (x64) is out now!
You can check the change log in our blog.
Go to Download page

User avatar
LRP
Posts: 132
Joined: 07 Apr 2008
Location: Switzerland
Has thanked: 81 times
Been thanked: 11 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby LRP » 02 Jan 2019

Thank you very much for providing the x64 AND the x32 version. Very appreciated!

User avatar
LRP
Posts: 132
Joined: 07 Apr 2008
Location: Switzerland
Has thanked: 81 times
Been thanked: 11 times

Data Playback?

Postby LRP » 06 Jan 2019

Dear Support

At my site, Data Playback do not work anymore.
What happens?

Thank you very much and kind regards

wegi
Posts: 119
Joined: 02 Jun 2009
Has thanked: 4 times
Been thanked: 12 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wegi » 06 Jan 2019

Tested the option "Format -> display entire series".
It does not work in release 4, as my last post in release 3.
i am not able to see 1 year of 5 minute data, i get an scrollbar.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: Data Playback?

Postby Henry MultiСharts » 07 Jan 2019

Dear Support

At my site, Data Playback do not work anymore.
What happens?

Thank you very much and kind regards
LRP, data playback works on our end. Please provide more details regarding the specific issue you have, or come to the live chat to demonstrate it remotely.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 07 Jan 2019

Tested the option "Format -> display entire series".
It does not work in release 4, as my last post in release 3.
i am not able to see 1 year of 5 minute data, i get an scrollbar.
wegi, as specified in Release 3 thread - we are currently checking it. I will let you know as soon as there are any updates.

automaton
Posts: 40
Joined: 04 Apr 2017
Has thanked: 2 times
Been thanked: 4 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby automaton » 07 Jan 2019

I got no update notifications for release 4. Checking for updates shows message “You have most current version”. I’m still running release 3.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 08 Jan 2019

Tested the option "Format -> display entire series".
It does not work in release 4, as my last post in release 3.
i am not able to see 1 year of 5 minute data, i get an scrollbar.
wegi, as specified in Release 3 thread - we are currently checking it. I will let you know as soon as there are any updates.
wegi, please follow the recommendation in this post.

automaton
Posts: 40
Joined: 04 Apr 2017
Has thanked: 2 times
Been thanked: 4 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby automaton » 10 Jan 2019

Am I the only one not getting a visual notification to update to release 4?

Checking for updates shows message “You have most current version”. I’m on version 3 currently

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 11 Jan 2019

I got no update notifications for release 4. Checking for updates shows message “You have most current version”. I’m still running release 3.
automaton, this build will become available in the "Check for Updates" feature next week. You can manually download it from our website now.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 11 Jan 2019

In MC12....... If you fat finger an optimization space input, you can no longer just press "cancel" and then reopen the "optimization settings" and start again. Now, it automatically stores whatever entry you input.

I feel pressing "x" or "cancel" should always revert any changes to previously saved. If MC insists on such behaviour, maybe only pressing the new "previous" tab should save changes?

If I make a mess of things, it a computing norm to be able to quickly roll back (ctrl+z, or cancel and don't save etc etc).

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 11 Jan 2019

In MC12's Optimizer I've got a conflicting message:
best or average.PNG
best or average.PNG (13.79 KiB) Viewed 1434 times
Is it best or average? Presumably not both??

Gaempi
Posts: 49
Joined: 27 Jul 2010
Location: Switzerland
Has thanked: 3 times
Been thanked: 4 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Gaempi » 14 Jan 2019

And please correct this:

1. In the optimization window: "Average fithness ..." =>> "Average fitness ...".

2. In the optimization report: the chaotic mix of the columns (% Profitable, input1, Avg Trade, input2, input3, Avg Bars in Loser, ...). If i have a lot of inputs to optimiz, in the optimization report the columns are very chaotic. As it was in MC 11 and it shoud be: First all the statistic columns and then all the input columns, but no mix of them.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 14 Jan 2019

I'm running tests on MC12r4 optimizer.

Can MC provide some clarifications on some of the new settings. At face value there seems to be an error somewhere with the core allocation feature.

Just as a reminder on the GA incremental behaviour:
viewtopic.php?t=51229

I think it is very cool to now be able to select the number of population members to be crossed over per generation:
set no. cores 2.PNG
set no. cores 2.PNG (53.67 KiB) Viewed 1356 times
If you remember generations can't be run in parallel, whereas backtests on offspring from the crossover process can be. So, from my understanding, if this is set to "2" the following should happen between generations (for an incremental GA):

1) 2 Children are generated (from the best individuals), upon which 2 backtests are run (1 per child)
2) These children are evaluated and placed into the population, replacing the worst. This creates our new generation
3) The above 2 steps are repeated.

Unfortunately these steps have to happen sequentially and not in parallel (otherwise there would be no evolutionary improvement, it would just be random).

So, I'm wondering why all 8 of my logical processors are kept busy and not just 2 of them, eventhough I've specified 2 in the settings?

Is the slider overriding my initial setting? Ie I've chosen "2", but thats being immediately replaced with "8" once the optimisation starts?
Core setting overridden.PNG
Core setting overridden.PNG (99.21 KiB) Viewed 1356 times
Thanks in advance.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 14 Jan 2019

In addition to my last post, I've found another error in the optimizer:

For an incremental GA, the number of tests to be conducted= Initial_population + (Generation*Children).

Children meaning the number of individuals to be crossed over between generations.

E.g. 200+(600*2) = 1400 runs.



The error I've uncovered:

If children>logical processors then the optimizer ignores the user defined value and reverts to the last setting used that was <=logical processors (or max logical processors if there was no last accepted setting). A pretty bizarre outcome.
No error/warning message notifies the user that their input wasn't accepted. Neither can I think of a reason why it can't be accepted......

If I want 80 children; then the optimiser simply has to perform 80 backtests before proceeding to the next generation, no? Does it matter if I have less than 80 logical processors? Surely it can be done sequentially (batches of 8, 10 cycles) until completed? The handling of this event is also pretty weird.

Let me know if I've got the wrong end of the stick.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 14 Jan 2019

I believe the Basic GA is also wrong (but could be mistaken). Number of runs seems wrong relative to Population and Generation sizes.

E.g. If I set 10 and 10 respectively; I get 125 runs.

Can't work that out. Is it an error?

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 14 Jan 2019

Do MC have access to the GA source code? Would they be prepared to share it, if so?

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 14 Jan 2019

wilkinsw, Gaempi, your inquiries are being analyzed.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 15 Jan 2019

I like how optimization reports are now automatically saved with a date and time stamp to a folder using the "Optimization Result Visualizer64" file format.

Is there a way to specify where they are saved? At the moment all reports are going to my Documents folder. Would be useful to be able to specify this.

It would also be great if this new "Optimization Result Visualizer64" functionality was extended to PT too. What I particularly like about it is that when you open a report you now have the "strategy info" icon in the top right which allows the user to check their results against the space that was searched. Very cool and would be awesome to have in portfolio trader.

Would also be great to have optimization reports load up with opened PT files; if the user has specified it in MC preferences (ie extend the same functionality that now exists within MC).

Obviously the new MC's GAO is now different to the older PT GAO. Just curious if extending the newer GAO to PT is on the roadmap?

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 15 Jan 2019

I've spent the day playing with the new MC12 WFA and can see a couple of improvements that would be great.

1) If we fail to find a "fit" In-Sample permutation; then skip the OOS run and proceed to next In-Sample optimisation.
OOS trades should be zero if not fit IS perms found 2.PNG
OOS trades should be zero if not fit IS perms found 2.PNG (114.18 KiB) Viewed 1289 times
In the above example my best returned fitness function value is zero. Ie nothing was found. When nothing is found; MC then chooses a random set of inputs instead for the OOS run. I would certainly not got live with a random set of inputs for my signal, if that signal could not be fitted to the in sample dataset.

I recommend adding a "greater than" value when the user is setting up the WFA. For the fitness function I used above that would be "0". But other users might want to specify values (e.g. If using Net Profit...... >$10,000).

2) It would be great if the user could specify which metrics to be included in the WFA report. Just like they can with a GAO. I can't speak for other users but I only use about 5 metrics. Being able to only choose 5 for the report would dramatically declutter it.

Thanks!

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 16 Jan 2019

I believe the Basic GA is also wrong (but could be mistaken). Number of runs seems wrong relative to Population and Generation sizes.

E.g. If I set 10 and 10 respectively; I get 125 runs.

Can't work that out. Is it an error?
I've just noticed that Portfolio Trader has a similar issue but for WFAs (both GAO and Brute Force)....... it is calculating too many runs to be backtested. Consequently it means, any form of PT WFA based optimization, takes significantly longer than the MC10 that I'm comparing it too.

I'll share some files with support so that they can reproduce.

The files will demonstrate how a Brute Force WFA (13 IS:OOS runs) search in a space of 64 permutations shows as needing 909 runs in MC12, vs what I make it: 13*(64+1) = 845 (same as MC10).

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 16 Jan 2019

wilkinsw, thanks for your feedback. After discussing it internally:
In MC12....... If you fat finger an optimization space input, you can no longer just press "cancel" and then reopen the "optimization settings" and start again. Now, it automatically stores whatever entry you input.
I feel pressing "x" or "cancel" should always revert any changes to previously saved. If MC insists on such behaviour, maybe only pressing the new "previous" tab should save changes?
If I make a mess of things, it a computing norm to be able to quickly roll back (ctrl+z, or cancel and don't save etc etc).
We will have it implemented in the next update.
In MC12's Optimizer I've got a conflicting message:
Is it best or average? Presumably not both??
We will have the message corrected in the next update.
I like how optimization reports are now automatically saved with a date and time stamp to a folder using the "Optimization Result Visualizer64" file format.
Is there a way to specify where they are saved? At the moment all reports are going to my Documents folder. Would be useful to be able to specify this.
We are going to add that option in the next major platform update.
It would also be great if this new "Optimization Result Visualizer64" functionality was extended to PT too. What I particularly like about it is that when you open a report you now have the "strategy info" icon in the top right which allows the user to check their results against the space that was searched. Very cool and would be awesome to have in portfolio trader
This is something that we consider adding to our roadmap for one of the future versions.
Would also be great to have optimization reports load up with opened PT files; if the user has specified it in MC preferences (ie extend the same functionality that now exists within MC). Obviously the new MC's GAO is now different to the older PT GAO. Just curious if extending the newer GAO to PT is on the roadmap?
This functionality may appear in one of the future versions.
I've spent the day playing with the new MC12 WFA and can see a couple of improvements that would be great.
1) If we fail to find a "fit" In-Sample permutation; then skip the OOS run and proceed to next In-Sample optimisation.
In the above example my best returned fitness function value is zero. Ie nothing was found. When nothing is found; MC then chooses a random set of inputs instead for the OOS run. I would certainly not got live with a random set of inputs for my signal, if that signal could not be fitted to the in sample dataset.
I recommend adding a "greater than" value when the user is setting up the WFA. For the fitness function I used above that would be "0". But other users might want to specify values (e.g. If using Net Profit...... >$10,000).
This functionality may appear in one of the future versions.
2) It would be great if the user could specify which metrics to be included in the WFA report. Just like they can with a GAO. I can't speak for other users but I only use about 5 metrics. Being able to only choose 5 for the report would dramatically declutter it. Thanks!
We'll try to fit it into the roadmap for the next major version.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 16 Jan 2019

I'm running tests on MC12r4 optimizer.

Can MC provide some clarifications on some of the new settings. At face value there seems to be an error somewhere with the core allocation feature.

Just as a reminder on the GA incremental behaviour:
viewtopic.php?t=51229

I think it is very cool to now be able to select the number of population members to be crossed over per generation:

If you remember generations can't be run in parallel, whereas backtests on offspring from the crossover process can be. So, from my understanding, if this is set to "2" the following should happen between generations (for an incremental GA):

1) 2 Children are generated (from the best individuals), upon which 2 backtests are run (1 per child)
2) These children are evaluated and placed into the population, replacing the worst. This creates our new generation
3) The above 2 steps are repeated.

Unfortunately these steps have to happen sequentially and not in parallel (otherwise there would be no evolutionary improvement, it would just be random).

So, I'm wondering why all 8 of my logical processors are kept busy and not just 2 of them, eventhough I've specified 2 in the settings?

Is the slider overriding my initial setting? Ie I've chosen "2", but thats being immediately replaced with "8" once the optimisation starts?
Core setting overridden.PNG

Thanks in advance.
In this specific case all cores are used only for the calculation of the first population of 136571 individuals.
As soon as the calculation of this population is complete, two cores will be used for further calculations, as the optimizer cannot move to the next generation before 2 children from the current generation are calculated.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 16 Jan 2019

In addition to my last post, I've found another error in the optimizer:
For an incremental GA, the number of tests to be conducted= Initial_population + (Generation*Children).
Children meaning the number of individuals to be crossed over between generations.
E.g. 200+(600*2) = 1400 runs.
The error I've uncovered:
If children>logical processors then the optimizer ignores the user defined value and reverts to the last setting used that was <=logical processors (or max logical processors if there was no last accepted setting). A pretty bizarre outcome.
No error/warning message notifies the user that their input wasn't accepted. Neither can I think of a reason why it can't be accepted......
If I want 80 children; then the optimiser simply has to perform 80 backtests before proceeding to the next generation, no? Does it matter if I have less than 80 logical processors? Surely it can be done sequentially (batches of 8, 10 cycles) until completed? The handling of this event is also pretty weird.
Let me know if I've got the wrong end of the stick.
The incremental algorithm produces worse results with an increase in the number of children.
Original algorithm implies only two children, therefore it shows the best convergence with two children.
The ability of increasing the number of the children up to the number of CPU cores was provided to get the results faster in the expense of quality.
Increasing the number of children more than the number of cores has no use since it won’t improve the speed and in the general case will produce worse results in comparison with the optimization with fewer children.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 16 Jan 2019

I believe the Basic GA is also wrong (but could be mistaken). Number of runs seems wrong relative to Population and Generation sizes.
E.g. If I set 10 and 10 respectively; I get 125 runs.
Can't work that out. Is it an error?
Do MC have access to the GA source code? Would they be prepared to share it, if so?
I've just noticed that Portfolio Trader has a similar issue but for WFAs (both GAO and Brute Force)....... it is calculating too many runs to be backtested. Consequently it means, any form of PT WFA based optimization, takes significantly longer than the MC10 that I'm comparing it too.
I'll share some files with support so that they can reproduce.
The files will demonstrate how a Brute Force WFA (13 IS:OOS runs) search in a space of 64 permutations shows as needing 909 runs in MC12, vs what I make it: 13*(64+1) = 845 (same as MC10).
Basic Genetic Optimization uses more parameters than just Population and Generation sizes.
The total number of simulations is not equal to Population * Generation.

Here is the formula for calculating the total number of simulations when using Basic GA.
Parameters used:
number of inputs enabled for optimization
MaxNumberofGenerations - Generation sizes
SetPopulationSize - Population size.
and the other parameters from the Optimization Settings window.

pmful = 1 - Math.Pow (1 - MutationProbability, Inputs);
resultSim = (SetPopulationSize + MaxNumberofGenerations * (1 + SetPopulationSize * (pmful + CrossoverProbability - pmful * CrossoverProbability))) + MaxNumberofGenerations;

Where:
Math.Pow (x, y) – the raising to power of x ^ y.
resultSim - the total number of simulations.
inputs - the number of inputs.
The remaining parameters from the Optimization Settings window.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 16 Jan 2019

And please correct this:
1. In the optimization window: "Average fithness ..." =>> "Average fitness ...".
It will be corrected in the next update.
2. In the optimization report: the chaotic mix of the columns (% Profitable, input1, Avg Trade, input2, input3, Avg Bars in Loser, ...). If i have a lot of inputs to optimiz, in the optimization report the columns are very chaotic. As it was in MC 11 and it shoud be: First all the statistic columns and then all the input columns, but no mix of them.
You can change the order of the columns by drag and drop. The last configured order is saved and used by all subsequent optimizations. In the report window -> Settings -> Columns you can disable the columns that you do not need and save the settings for the subsequent reports by enabling the checkbox "Use as Default" and clicking OK.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 16 Jan 2019

I believe the Basic GA is also wrong (but could be mistaken). Number of runs seems wrong relative to Population and Generation sizes.
E.g. If I set 10 and 10 respectively; I get 125 runs.
Can't work that out. Is it an error?
Do MC have access to the GA source code? Would they be prepared to share it, if so?
I've just noticed that Portfolio Trader has a similar issue but for WFAs (both GAO and Brute Force)....... it is calculating too many runs to be backtested. Consequently it means, any form of PT WFA based optimization, takes significantly longer than the MC10 that I'm comparing it too.
I'll share some files with support so that they can reproduce.
The files will demonstrate how a Brute Force WFA (13 IS:OOS runs) search in a space of 64 permutations shows as needing 909 runs in MC12, vs what I make it: 13*(64+1) = 845 (same as MC10).
Basic Genetic Optimization uses more parameters than just Population and Generation sizes.
The total number of simulations is not equal to Population * Generation.

Here is the formula for calculating the total number of simulations when using Basic GA.
Parameters used:
number of inputs enabled for optimization
MaxNumberofGenerations - Generation sizes
SetPopulationSize - Population size.
and the other parameters from the Optimization Settings window.

pmful = 1 - Math.Pow (1 - MutationProbability, Inputs);
resultSim = (SetPopulationSize + MaxNumberofGenerations * (1 + SetPopulationSize * (pmful + CrossoverProbability - pmful * CrossoverProbability))) + MaxNumberofGenerations;

Where:
Math.Pow (x, y) – the raising to power of x ^ y.
resultSim - the total number of simulations.
inputs - the number of inputs.
The remaining parameters from the Optimization Settings window.
Thanks for the detailed replies.

That just leaves the discrepancy regarding MC12 PT WFA (exhaustive) excessive calculation runs. In the example I sent you, MC12 performs 909 calculations and MC10 845 calculations. Unsurprisingly, MC12 is slower too.

hughesfleming
Posts: 231
Joined: 22 Apr 2014
Has thanked: 57 times
Been thanked: 58 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby hughesfleming » 17 Jan 2019

Thank you @wilkinsw for doing these tests.

Alex

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 17 Jan 2019

MC, I think you forgot to have a look and change this:
viewtopic.php?f=1&t=51488&p=129983&hili ... rs#p129983
still doesn't look correct or make sense.PNG
still doesn't look correct or make sense.PNG (16.91 KiB) Viewed 1224 times
The purpose of these two boxes is to spot a "rejectable" stop order (rejectable because last traded price is worse than the trigger price, implying its intention for immediate execution as an aggressor order, but making it rejectable by the exchange) and convert it to an aggressor order for immediate execution.

1) You only need 1 box here, not two. Have something like: "Convert stopmarket/limit orders intended for immediate execution/trigger to market/limit orders".

2) Can I ask who is detecting the rejectable stop orders? If it is on MC's end can you please make sure that the exchange rejection reason matches MC's definition of rejectable. For example as a general rule, all futures exchanges use "last traded" and check that value against the stop trigger price. It has to be a c.100% match otherwise users will suffer false stop triggers (instead of having a stop working, untriggered at the exchange). If there are CQG FX users then maybe you want to insert a second box that they can tick too to make stop conversion quote relative???

I really want to use this feature as it will save me coding exception handlers (for pre-empting rejected orders) but the two options shown are gobbledygook and I've come against false synthetic stop conversions in MC before.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 17 Jan 2019

Thanks for the detailed replies. That just leaves the discrepancy regarding MC12 PT WFA (exhaustive) excessive calculation runs. In the example I sent you, MC12 performs 909 calculations and MC10 845 calculations. Unsurprisingly, MC12 is slower too.
wilkinsw,

After the last run the optimization is done once again on a data series equal to IS (or the complete data range for Anchored). The last WFA bar is matching the last bar of the data series. As the result of this optimization the Recommended set of inputs is displayed.
Your chart is the OOS for the recommended parameters.
When you apply them to signals – they are backtested with these recommended inputs on your chart.
There was no such functionality in MC 10, hence the difference in the number of the calculations:
MC 10 (64 + 1) * 13 = 845
MC 12 (64 + 1) * 13 + 64 = 909

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 18 Jan 2019

Thanks for the detailed replies. That just leaves the discrepancy regarding MC12 PT WFA (exhaustive) excessive calculation runs. In the example I sent you, MC12 performs 909 calculations and MC10 845 calculations. Unsurprisingly, MC12 is slower too.
wilkinsw,

After the last run the optimization is done once again on a data series equal to IS (or the complete data range for Anchored). The last WFA bar is matching the last bar of the data series. As the result of this optimization the Recommended set of inputs is displayed.
Your chart is the OOS for the recommended parameters.
When you apply them to signals – they are backtested with these recommended inputs on your chart.
There was no such functionality in MC 10, hence the difference in the number of the calculations:
MC 10 (64 + 1) * 13 = 845
MC 12 (64 + 1) * 13 + 64 = 909
Hi Henry,

I'm still getting PT WFAs with number of calculations that don't equate to your logic.

I'll send the files to you now.

For example:
PT WFA GA incremental: 17 IS:00S runs (18 IS runs if MC12).
Pop=8, Gen=1, logical processors=8

That should be:
MC10 289 calculations
MC12 305 calculations

However, it's stating:
MC12 526 calculations
FYI MC10 is correct (289).

What's wierd, is that when MC12 roughly reaches the max calculations it should be doing (near 305), the progress bar then accelerates forward. Maybe it's then skipping the "phantom" calculations and proceeding to the final IS run??

In any event, I'd appreciate you guys taking a look. Hopefully it's just a bug in the calculation of cycles and doesn't actually effect what the WFA ends up doing. Would be good to not have to manually work out an eta for WFA completion.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 18 Jan 2019

I've uncovered that the above issue seems to only effect PT anchored WFAs. Non-anchored generate rational numbers for number of calculations to be conducted.

However, I've now also discovered a discrepancy in how PT calculates the size of windows in anchored vs non-anchored.

In the example below, they are both 1000 day in sample and 125 out of sample. The left results are non-anchored and right are anchored. Total runs should be the same for both, but aren't. The first IS:OOS run should have the same dates but don't. 1000 days in anchored translates to very slightly under 4 years (as in 1000/252 trading days). And 125 days is roughly 6 months. But, for some reason, means something else if non-anchored.

I can't work out why turning off "anchored" is creating such different behaviour. And why can't an anchored WFA correctly calculate the number of calculations to be carried out? Might the two aberrations be related?
very strange behaviour of anch vs non anch.PNG
very strange behaviour of anch vs non anch.PNG (805.33 KiB) Viewed 1190 times

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 18 Jan 2019

Also,

In MC12 PT GAO settings: If you choose "basic", it won't auto populate the population and generation fields, like it does in MC10 and like it does in MC12's native GAO. Instead it sticks with the auto-populated values that were generated for the default incremental GA.

wilkinsw
Posts: 529
Joined: 21 Apr 2013
Has thanked: 135 times
Been thanked: 86 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby wilkinsw » 19 Jan 2019

I'm not convinced the WFA robustness measure is on point.

I think it is a great idea for a feature.

Might I suggest when the user clicks on "Robustness: PASS/FAIL" in the bottom right that the window that pops up, not only shows them the robustness settings currently in place, but also shows individual PASS/FAIL status for each measure.

I am yet to go through this feature with a fine tooth comb. I've just completed a PT WFA and I've deduced that robustness failed due to average profit factor being less than X. I presume that means average OOS profit factor?

However, when calculating the average OOS PF in excel, i have a figure much higher than the robustness threshold. So no idea why it is failing on that measure. It could be a bug maybe?

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 23 Jan 2019


User avatar
Anna MultiCharts
Posts: 415
Joined: 14 Jul 2017
Has thanked: 30 times
Been thanked: 104 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Anna MultiCharts » 31 Jan 2019

Hello, wilkinsw!

RE: stop and stop-limit orders conversion.
We’ve exchanged multiple emails on this matter last year and arrived to consensus that we’ll perform the conversion this way:
viewtopic.php?t=51308#p128897
These conversion principles were introduced in MultiCharts 11 Release 9. Have you already tried using these options? Are they working as described in the above specified forum post? If not, please describe what’s wrong in more detail and provide some illustrations for this. (maybe it would be more convenient to start another forum thread in this case)
As for combining options into one – I’m afraid most likely we won’t be able to do that as other users might be already accustomed to this setup.
Can I ask who is detecting the rejectable stop orders?
– conversion takes place upon order placement, and MultiCharts makes the decision about conversion.
If there are CQG FX users then maybe you want to insert a second box that they can tick too to make stop conversion quote relative???
– please elaborate.

User avatar
Henry MultiСharts
Posts: 9081
Joined: 25 Aug 2011
Has thanked: 1257 times
Been thanked: 2920 times

Re: MULTICHARTS 12.0 RELEASE 4

Postby Henry MultiСharts » 28 Feb 2019

In MC12 PT GAO settings: If you choose "basic", it won't auto populate the population and generation fields, like it does in MC10 and like it does in MC12's native GAO. Instead it sticks with the auto-populated values that were generated for the default incremental GA.
This is something that will be corrected when Portfolio Trader is switched to using the new optimization GUI (like in MultiCharts).
Might I suggest when the user clicks on "Robustness: PASS/FAIL" in the bottom right that the window that pops up, not only shows them the robustness settings currently in place, but also shows individual PASS/FAIL status for each measure.
This is something we are going to improve in MultiCharts 13.
However, when calculating the average OOS PF in excel, i have a figure much higher than the robustness threshold. So no idea why it is failing on that measure. It could be a bug maybe?
If this is something that still can be replicated on your end, please send me the workspace, data, signal for replicating it on our end.


Return to “MultiCharts”