[Push] Input stack (from Psh)

PerPlex Ed edperplex at yahoo.com
Mon Jun 28 18:08:16 EDT 2010


I fixed some bug in the evolution code and today it found approximative (0.1 rmsre) solutions with both x^2 and x after 300 generations. Previously I only found x^2.

(((((0.150111978908561) 0.528050625819725 Integer.* Float.*)) Float.+ Float.Swap (Float.Rot Integer.- Integer.Max (((((Float.< Integer.Min Float.Dup (((Float.Rot) (1.0142690849058 Float./ Integer.+ Integer.+ Integer.-) Integer.Max Float.*))) 1.94731824936981 Float.+)) 1.64833089095734 Float.+)) Float./) Float.=) Float.+ Integer.% (Float.+ 1.52025841305038 1.94731824936981 Float.Min Float.*) Integer.Swap)

(50 points) which can be simplified to

(0.150111978908561 0.528050625819725 Float.* Float.+ Float.Dup 1.0142690849058 Float./ Float.* 1.94731824936981 Float.+ 1.64833089095734 Float.+ 1.52025841305038 1.94731824936981 Float.Min Float.*)

(17 points)

This is without "inputs instructions" (I didn't implement any kind of them) and with ephemeral random floats limited to [0.0, 2.0]. No simplification is used during the evolution yet. Env, Name, Boolean, Exec and Code are "disabled".


PerPlexEd said:
> http://img375.imageshack.us/img375/922/myfirstpushsymbolicregr.jpg
> But it took 800 generations and a good amount of luck to get that. :)


Lee said:
> PS, that last run (with the changed ephemeral randoms and an input instruction) the error was down to 0.0276408490944533 at generation 98.


      



More information about the Push mailing list