Math Vector Library 1.1 Machine Learning Extension
Programming Reference Manual Fortran
Version 1.1 DD-00004-111

4.30 vmlsmhtd - SMHT activation (first derivative)

subroutine vmlsmhtd(n, y, incy, a, b, c, d, x, incx) 
subroutine vmlsmhtdf(n, y, incy, a, b, c, d, x, incx) 
 
subroutine vmlsmhtd_64(n, y, incy, a, b, c, d, x, incx) 
subroutine vmlsmhtdf_64(n, y, incy, a, b, c, d, x, incx)

Given an input vector x  , a result vector y  and constants λ and α this function computes the first derivative of the Soboleva modified hyperbolic tangent (SMHT) activation function of the values in the x  vector and stores the result in the y  vector.

 . aeax+be-bx         cecx-de-dx
y= -ecx+e-dx---smht(x)-ecx+-e-dx--

4.30.1 Parameters

N - INTEGER

ENTRY: Number of elements of x  and y  .
CONSTRAINT: n ≥1  .

Y - ARRAY OF REAL

EXIT: Result vector y  .
CONSTRAINT: Must contain n× incy  elements.
CONSTRAINT: Must not overlap with array x  .

INCY - INTEGER

ENTRY: Stride for the vector y  .
CONSTRAINT: incy̸= 0  .
BEHAVIOR: A negative stride will traverse the array in reverse.

A - REAL

ENTRY: Constant a .

B - REAL

ENTRY: Constant b .

C - REAL

ENTRY: Constant c .

D - REAL

ENTRY: Constant d .

X - ARRAY OF REAL

ENTRY: Input vector x  .
CONSTRAINT: Must contain n× incx  elements.
CONSTRAINT: Must not overlap with array y  .

INCX - INTEGER

ENTRY: Stride for the vector x  .
CONSTRAINT: incx̸= 0  .
BEHAVIOR: A negative stride will traverse the array in reverse.