Two ideas, how to improve Builder in the future

[Michal Rydlo]

Hello,

as I use Adaptrade Builder for a couple of weeks/months, I realised
that I miss two functions which I believe wouldn't be so hard to add,
but help users a lot.

1. When the building of strategies is done, I have the option the sort
the perfomance In-sample (which has no big sense alone) and Out-of-
sample. It would be great, if in the summery would be the column (or
the separate list) where I could see the performace (for example total
profit, profit per trade or anything else) for the whole period (IS
+OOS). The reason is, that in some cases, the strategy can have the
great performace OOS, but is losing IS (which is unusable as well). It
would be great if I could filter the strategies by the total results IS
+OOS (it would be even better if I could set different weights for
results IS and OOS).

2. As many users I am limited by the size of memory when building (the
best I can do is 3 generations and population of 5000 - and it takes
about 2 hours). But, when I go out, I would like to use also the rest
of time after this one build, but not losing the good strategies (even
if they do not fit the conditions for not being re-built). Would it
possible let the Builder work in this way: Build 2 gen. and 5000 pop.
==> after this, save the 20 best strategies according to total profit
OOS ==> repeat the build (you still have those 20 strategies saved
beside and do not change them anymore) ==> after, save another 20 best
str. of the new build ==> repeat this for example 5 times (until I get
back home from the work :) ) ==> after all those builds, show me all
the saved strategie (5x20).
Thanks to this, users would be able to generate much more strategies
with pretty good results on which we could do more tests after. The
strategies wouldn't be correlating much as if I would only add
generations.

I know you are doing a hard work on the Builder all the time, but if
you could consider add these function in some of the next releases (as
I said, I think it wouldn't be so complicated as you are not changing
the build algorithm at all), it would be great.

Thanks and have a nice day.

Michal Rydlo

[mandelmus]

+1 vote on the 2nd idea.

 

 

As far as the 1st idea...

The reason is, that in some cases, the strategy can have the great performace OOS, but is losing IS (which is unusable as well)

I used to think that way, but not so much anymore.  My new approach is to have Builder create a very large population of strategies over a relatively short IS period, with <3 generations, then "Evaluate All" over a larger OOS period (weeks-months-years).  If the IS period is relatively short and is experiencing a loss, then it's possible that the strategy just didn't work well over the IS period.  If the same strategy is profitable over the larger OOS period, why throw out that strategy just due to the IS performance?  If you use a large enough OOS time frame and the results are profitable, just consider yourself lucky that Builder found a profitable strategy over that period.