Digamma function .
Implemented as a composite of engine primitives so autograd flows
through naturally, and its gradient is the trigamma function. For
x >= 0 the argument is shifted by 8 via
and the
asymptotic Bernoulli series is evaluated there; for x < 0 the
reflection formula
maps the argument onto the right half.
Parameters
inputTensorReal input. The negative integers and
-inf are poles and give
NaN; psi(+0) is -inf and psi(-0) is +inf.Returns
TensorSame shape as input.
See Also
lgamma— itself.