System building from blocks (units).

[VladGor]

System building from blocks (units).
(for discussing)

With the new optimizer the ability of strategy building from several signal became available, as it is available in the Omega Strategy Builder.

The idea of the approach.

Strategy is built from the different separate signals, implied in a special way. The difference from the regular signal (as it is in Omega) is that GO signal has in its beginning the block code which defines optimizable signal parameters (genes) and sample of the parameters values from the optimizer.

At the same time signal might not has any external parameters (operator Input).

Optimization is controlled from the separate block (signal), which is also included in strategy. This block is universal and might control the optimization with any sample of signals.

The control block has at least Gen parameter. Also it might have parameters for size population defining, for operating mode defining and so on.

This block organizes optimization for every signal. In this block OOS as well as WFO might be implemented in the future.

It is possible to include in the release standard block configuration so a client will just have to define his own signals.

The block is might included fitness calculation by using standard algorithms. Also it’s possible to implement an additional block (signal) for non-standard fitness calculation and also include this block into the strategy.

The example of the control block TS.GO.Manager

[kshultz]

I am still unclear as to the purpose of the variable: 'show' or 'showind'
" { If this is the last path, shows results for Ind = ShowInd;
Else get the next candidate Ind = 0; }
Ind = Iff(LastRun = 1,ShowInd,0); "

What is the impact or difference between using different values for showind? using 1 or using say, 10 with Gen = 1 to 1000?

Thank you

[VladGor]

[kshultz]

This does not really answer my question. Please show an example of code using two different signals, for example the moving average cross over you show as an example and the addition of another signal, like an RSI, testing whether or not one is better than another combined or separate, for different inputs ( example for RSI might be inputs for RSILength, OverBoughtLevel, OverSold Level). In other words, MA Cross might be better with RSI above OB or below OS or RSI might be better than MA by itself. Might even want to test several signals with different input values. Thank you ks

[Mak]

If we correctly understood your question then example is below.
We call it structural optimization that is optimization not only on parameters but also on system content.

For that in every signal one additional gene is added with two positions (On/Off) that manages the state of signal:

[kshultz]

How to combine 2 signals in one GO evaluation?

[Mak]

Well ...
It's not a TS.GO problem.
It's EasyLanguage programming problem ..

The next code have the same result,
but allow combine any number of signals.

[kshultz]

Thank you, I think that will work. I will try it. ken

[krc]

The following code actually runs with the optimizer but I'd like to be able to abstract the method and use something like the TSGO.Manager to allow a more orderly approach. The difficulty is that I want all three rules to be valid in order to enter a position. Today the TSGO.Manager passes this decision to the children. I'd also like to switch between (optimize) the entry type (market, limit, stop, etc.)

[krc]

I think the question is about using Chromosomes correctly. The default translation of the above code through the 'Code Converter' is not optimal. I expect that at least each 'Rule' should have its own chromosome.

[VladGor]

We didn't quite understand your question.
Do you want us to help you to modify code of your system for TSGO so that every rule has separate chromosome?

Regarding convertor.
Convertor is made to simplify work or users.
More experienced users can add new chromosomes by themselves after conversion.

[krc]

Let me rephrase my questions:

If I have a complex strategy (like the one above), how do I determine the right number of chromosomes to use?

Are there any other factors that I should consider?

With this particular strategy my guess is three chromosomes. Is this correct? (and it would be six if longs and shorts were not discrete)

Is there any easy way to utilize TSGO.Manager in a way that the actual order is not given until multiple strategies vote that an order is given? Maybe using TSGO.GO.Set in the child strategies and TSGO.GO.Get in the actual order placing strategy.

Or maybe I should have 30 strategies / chromosomes each using an additional gene with On/Off?

Would that allow up to 30 strategies to vote on a given order?

Or maybe I can't do this in a Genetic Optimizer and need a Neural Net.

[krc]

I think 10 systems with the (on-off) gene might do what I am trying to accomplish.

At least it should be interesting...

Sorry for any confusion...

[VladGor]

[Hannibal]

Hi; I just got your toolset. I would like to request clarification of how it seems to work;

1. A gene is simply a parameter that you wish to test (optimise) and as long as it is numeric, it is usable.

2. A chromosome is used to group specific genes together so (for example) in a strategy with 20 different setups, you would logically put all of the genes for each setup into their own unique chromosome - that way the system wastes less time testing non-correlated combinations between chromosomes.

3. In hierarchal systems such as a system that has algorithms that determine biases for a given market environment, and then has setups that are unique to each type of environment, you should use a chromosome for each market environment algorithm (assuming more than one) and then individual unique chromosomes for each "sub-setup."

4. If you set the GOMode to 1, and place the Population input ABOVE the Gen input, then you can effectively extend the possible iterations (generations) from 100 up to 100,000 by optimising (in Tradestation) on Population = 1 to 100 AND Gen = 1 to 1000; Since Gen is second in line, TS will set Population to 1 (this is an extreme example, I would probably start population at 20, or else I would add a bit of code to allow me to count DOWN instead of UP.) and then it will cycle through all values of Gen before incrementing Population and restarting. This should effectively cause you to analyse each result set as the basis for the next analysis until you reach the last Population value. Actually, I suppose you could even just introduce a dummy input (unused except for being declared) and use that as your "recycling" counter for reiterating infinitely (if you wished).

Comments, and especially corrections please!

Also, is there a more detailed document on how to use this toolset - the stuff on the site is rather light in terms of being comprehensive.

Marc