System building from blocks (units).

[VladGor]

[Hannibal]

Let me be more clear about question 4:

I have a system that (currently) has 21 chromosomes, each with anywhere from 2 to 15 parameters. There are 3 chromosomes that focus exclusively on general market biases and the rest are actual setups (signals) tied to entry and exit code.

Overall that's roughly 150 genes each with an average of 25 possible values yielding a rather large number of possible permutations.

Given that my population is limited to 1000 and the number of generations is limited to 1000 - I need to perform several test runs in order to be comfortable that I have tried a reasonable subset of the available permutations.

First off: In performing "serial" test runs, should I set GOMode to 1 or 2? Which yields the best results or the fastest run time?

Secondly; All other parameters being correctly set, if I use an extra "dummy" input and have it count say from 1 to 5 in the TS optimisation engine while simultaneously optimising Gen from 1 to 1000, will I get the same results as if I had run a simple Tradestation optimisation of Gen from 1 to 1000, and then manually restarted the whole process again (with GOMode set to 1 or 2), repeating this process 5 times?

Thanks for your help!

Marc

[VladGor]

[Hannibal]

Thanks Vlad;

So if I understand it correctly -

GOMode = 1 is used when either the underlying system has changed, or the underlying data (or data window) has changed since it takes the successful result population and recalculates the system using those inputs in order to validate them. In this case, Gen must be set to at least the population size, since it must first recalc each of the existing entities in the population before taking on any new generations.

GOMode = 2 would be the default if we were simply iterating through the same system and data, since recalc would be pointless.

You also mentioned setting Gen to interate 2000 times - I thought we were limited to only 1000 generations - is this not true? If not, is there a limitation to the number of Generations and if so, what is it?

[VladGor]

[Hannibal]

Hello Vlad;

I have been running now for a few weeks and am getting great results - thanks for this great tool!

I have found the SystemQuality and TS.Fitness functions already in the forums, but I was wondering whether you had any other algorithms that you find useful - if so, would you consider posting them here?

In addition, I would like to post some pseudocode that I am now using to define fitness in a two tiered manner. My theory, underlying this concept, is that it is better to give unfit solutions a low weighting rather than a negative weighting, in order to allow "the best of the worst" to continue being considered within the population until such time as they are replaced by better solutions.

The algorithm works basically like this:

1) Using target values for the performance statistics that you wish to test, you normalise them such that meeting or exceeding the target generates a value of 1 and falling short generates a fraction representing how close the solution got to the target.

2) If all of the targets are met, then the fitness algorithm switches to evaluating the performance using relative merit as defined by a second "open ended" calculation that has a minumum value > 1.

I'm only including pseudocode here because it would be up to the user to determine which parameters they wished to test, and how many, as well as what they wished to use as a step 2 fitness test.

[VladGor]

[mjo]

I am interested in using the Block Manager concept to combine and test signals. I am having a problem with the TS.GO.Manager code.

To make sure I understand the concept, so far I have created three separate Easylanguage strategies: TS.GO.Manager, TS.GO.MA4 and TS.GO.ATR. I used the code blocks that were provided in this forum to create these files. When I apply these three strategies to a chart and run an optimization, the resulting trade names and "buy sell" lines do not show up on the chart. I can open the TS Genetic Viewer and see that the optimization has run and there are results. These results also show up on the TradeStation Optimization, but the Strategy Performance Report says "no trades were generated."

As a check, I incorporated the code from the three above strategies into a single strategy file, making the appropriate edits to remove duplicative terms. When I run this single strategy file, the trades do show up on the chart and in the Strategy Performance Report. While this is a "fix" it defeats the purpose of using the TS.Go.Manager as a "system builder" tool that can test multiple signals independently, as I understand the concept.

FYI, the "getstrategyname" is no longer a valid reserved word in Easylanguage, but this is not the problem.

[mjo]

After a few weeks of using the TSGO I can answer my own question.

The code above for the TS.GO.Manager is missing a reference to the ShowInd function that writes the final solution to the chart.

The code in the Manager should have a line like:

Ind=Iff(LastRun=1,ShowInd,0);