strange behavior of genetics

Voted for this task.

I guess it runs tests on all data in the moment of testing which I guess is fine in case you have an oos filter. If you do not have an oos filter you can turn off oos....

I imagine you will get the same results if you do that test with random generator too.... or really with any task. The test runs first then filters are checked after so it needs to run a full test first in case you have an oos filter.

this behavior is not acceptable - the slowdown occurs right after the start button is clicked, so the islands are starting to fill with initial population, which cant be filtered by any OOS data. Initial population must be choosen only from IS data

after that genetics will start and make the filtering by IS or OOS ranking - this has nothing to do whats happening in b130

so by my opinion added OOS if IS stays the same cant result to slow down the building of initial population

"so the islands are starting to fill with initial population, which cant be filtered by any OOS data. Initial"

Yes actually init pop can currently be filtered by OOS data. Would anyone ever WANT to? probably not. I don't know why this is possible but it is.

It should probably be ISV not OOS here but sadly it is really OOS. According to my testing ISV is included in the IS part so there's no way to separate IST and ISV at the moment in the init pop filters <--- this is a problem.

To me it seems most logical to not allow OOS in builder data, instead only allow IST and ISV. Then also make the init pop filters able to filter by those. It would be less confusing and more useful.

In the meantime can't you just run builder without OOS then do a retest after which includes the oos?

Voted for this task.

In any case, build task and retest task does only one whole main data test first for each strategy then checks filters after. (EDIT: It's true that checks against the global ranking filters are not done until the islands are full BUT when it checks the final global ranking filters it just uses the test from memory that's already been done during building, I'm 99% sure.)

 It does not do two separate tests I.E. test in sample in one test then check filters then do OOS test if passes filters (You can accomplish this though by using two separate tasks)

Voted for this task.

Voted for this task.

Voted for this task.

The more I think about it the more I agree with Hankeys. It would be nice to check in-sample filters before proceeding to do an OOS test.

Voted for this task.

Voted for this task.

Voted for this task.

Voted for this task.

the explanation is simple - you are backtesting the strategy on 4 years in faster case and 16 years in slower case.

Strategy is always backtested on full configured data range, stats for IS / OOS parts are computed only after that.

I understand your point that to speed it up it could backtest strategy only on IS part, but it is not that simple. 

You can have multiple IS parts scattered in the range, and you can have conditions anywhere that count with Full sample or OOS.

Voted for this task.