SQ Xのアップデート

ここからSQ Xのアップデートやニュースについて話し合うことができます。.

このトピックは「SQ4 Early Preview」から名称変更されました。古い、もはや関連のない投稿は削除され、アーカイブされます。.

ポイントは、古い、関連性のない内容からフォーラムを整理することです。.

Hey Mark a quick update on release 5 – it has been running for around 26 hours now and hasn’t crashed.  Thumbs up to this!

ive set up a filter to remove strategies where the ret/dd of monte carlo tests falls to <=50% of orginal ret/dd. Its still letting through strategies that don’t meet this criteria. Ive included an image so you can check if I set it up correctly. if its a bug so I can report it. thank you

This RC5  is a good step. It looks that we are slowly heading to final version.

can i get some info about the genetics and things behind it – how it works, what meant this setting…

what means islands – this wasnt in SQ3, so i dont know how it works

what is the best settings by the meaning of the community?

because with this settings – it will generate near 10000 strs for initial population and after 8 hours the fitness still holding and new initial population hasnt been generated

does the restart function if the generation stagnates really works?

if i set 10/10/10 it will generate 100 strs as an initial population and the restartinf seems to work – because after some time in the statistics was 400 strs as an initial population

but the graph of fitness function still holding as a straight line…i think there is a problem

Islands are a new feature in SQ4. Running genetic evolution on multiple islands means that there is an independent evolution run on every island, with ocassional migration of soem strategies from one island to another.

Its advantage is better diversity of strategies and that genetic evolution is not stucked in one locale maximum, as each island is independent and starts with its own initial population.

Its disadvantage is that you have to multiply the population by number of islands, because random initial population will be generated for every island.

If you set islands=1 then it behaves as same as in SQ3. I think the reasonable number is 5-10 islands, I don’t see the point of having more than 20 islands.

There is one more thing that influences the size of initial population – it is decimation coefficient. If you set it to something bigger than 1 it means it will create bigger initial population, and the best candidates are chosen.

For example, your population is 10, you set decimation to 3.

It will generate 3 x 10 = 30 filtered strategies, and top 10 from them will be selected as initial population, the rest will be thrown away.

You can see that the number of strategies in initial population can grow quickly. If you have population=50, decimation=3, islands=10 then SQ has to generate 50 x 3 x 10 = 1500 strategies that pass the initial filter only to generate initial population.

thx Mark for explanation…and what about setting to filter initial population – its better by your opinion to set something like that, “netprofit higher 0” – so only profitable strategies going to initial population – or keep it simple and filter nothing and do the genetics does its job and find the best results?

I’d say it is preferable to start with strategies that already have Net profit > 0, but it is a question.

It might be interesting to make some study about it.

I have been running 2 sessions like this for a week actually (on the same machine).If You set net profit to  0 it will in the same time same go thru 100% + more randome generated strategies then if You have it set to -5000. The end result seems to be that it finds 20 times more strategies using the same ranking if it is set to  -5000 which proves that Genetic Evo is far more superior then randome generation. However i have pretty nice result setting a win% instead of net profit . This would maybe be more logical then net profit since We are looking for strategies that have an edge on the market.

SQ4 is still missing improver. Will this be implemented ? Then we can search for strategies with fixed same size stop and target with a win% above 50 which would indicate that the entry had an edge on the market and then we could move them to improver to find better exits.

Anotherway to do that could be to save all exit combinations in the strategy but deactivated and then have WFA to activate them and create a new strategy thru Walkforward optimasation. But it would not work with indicator based Exits but be better then nothing.

Ran some test. SQ4 is very, very good in finding strategies with and Entry edge. I have 300 strategies in data bank after just a couple of hours and that’s on 11 years OOS tested on additional markets. With SQ3 it took weeks of generation. WOW ! We need improver function added or WFA activation of dormant exits ! Or both , he he

Testing different exits only on a strategy should be possible using GPU instead of CPU. Just generate a file with 1 billion predefined exit values and let a bunch of GPU’s figure it out..

mabi – Improver is in SQ4, you have to choose the right option in What to build setting.