Re: Differentiable Forth

And some other examples for gradient calculation ( 2 variables), with
different implementations (fstack juggling, using fvalues, using
floating point arrays (2 variants))
The code begins here:
cr .( gradient)
: gradient ( f: x y -- df/dx df/dy)
     ' func !
     fover fover ( f: x y x y)
     fswap f>dl_d ( f: x y y x 1)
     frot  f>dl   ( f: x y x 1 y 0)
     func @ execute fnip ( f: x y df/dx)
     frot frot ( f: df/dx x y)
     fswap f>dl ( f: df/dx y x 0)
     frot f>dl_d ( f: df/dx x 0 y 1)
     func @ execute fnip ( f: df/dx df/dy)
;
cr 2e 3e gradient dl_f10() fswap cr f. cr f.
cr cr
cr .( gradient using values)
0e fvalue x
0e fvalue y
: grad ( f: x y -- df/dx df/dy)
    ' dup
    to y to x
    x f>dl_d y f>dl   execute fnip
    x f>dl   y f>dl_d execute fnip
;
(  defined)
cr 2e 3e grad dl_f10() fswap cr f. cr f.
cr cr
cr .( gradient using arrays)
create xy 2 floats allot
xy 2 floats erase
: grad1 ( f: x y -- df/dx df/dy)
    ' dup
    xy float+ f!
    xy f!
    xy f@ f>dl_d xy float+ f@ f>dl   execute fnip
    xy f@ f>dl   xy float+ f@ f>dl_d execute fnip
;
(  defined)
cr 2e 3e grad1 dl_f10() fswap cr f. cr f.
cr cr
cr .( gradient using arrays)
create xy2 2 floats allot
xy2 2 floats erase
: to_xy2  xy2 float+ f! xy2 f! ;
: df/dx xy2 f@ f>dl_d xy2 float+ f@ f>dl   execute fnip ;
: df/dy xy2 f@ f>dl   xy2 float+ f@ f>dl_d execute fnip ;
: grad2 ( f: x y -- df/dx df/dy)
    ' dup
    to_xy2
    df/dx
    df/dy
;
(  defined)
cr 2e 3e grad2 dl_f10() fswap cr f. cr f.
cr cr
cr cr cr
The code ends here
This code is the continuity of the previous code in my last post.
Ahmed