def compute(self, arg):
r"""
Compute the hyperbolic tangent function value for a specified argument.
:param arg: argument to the hyperbolic tangent function.
:type arg: float
:returns: the hyperbolic tangent function value.
:rtype: float
EXAMPLES:
Once an instance is created, the activation function values are
obtainted through invocation of :meth:`compute`:
>>> from yaplf.utility.activation import \
... HyperbolicTangentActivationFunction
>>> f = HyperbolicTangentActivationFunction()
>>> f.compute(0)
0.0
>>> f.compute(1)
0.46211715726000974
The higher the value for :obj:`beta`, the more a sigmoidal function
approximates a step function (i.e., that constantly equal
to ``-1`` when its argument is negative, constantly equal to ``1``
otherwise):
>>> f = HyperbolicTangentActivationFunction(10)
>>> f.compute(0)
0.0
>>> f.compute(1)
0.99990920426259511
AUTHORS:
- Dario Malchiodi (2010-04-02)
"""
return (exp(self.beta * arg) - 1) / (exp(self.beta * arg) + 1)
def compute(self, arg):
r"""
Compute the sigmoid function value for a specified argument.
:param arg: argument to the sigmoid function.
:type arg: float
:returns: the sigmoid function value.
:rtype: float
EXAMPLES:
Once an instance is created, the activation function values are
obtainted through invocation of the :meth;`compute` method:
>>> from yaplf.utility.activation import SigmoidActivationFunction
>>> f = SigmoidActivationFunction()
>>> f.compute(0)
0.5
>>> f.compute(1)
0.7310585786300049
The higher the value for :obj:`beta`, the more a sigmoidal function
approximates the Heaviside step function (i.e., that constantly equal
to ``0`` when its argument is negative, constantly equal to ``1``
otherwise):
>>> f = SigmoidActivationFunction(10)
>>> f.compute(0)
0.5
>>> f.compute(1)
0.99995460213129761
AUTHORS:
- Dario Malchiodi (2010-02-22)
"""
return 1 / (1 + exp(-1 * self.beta * arg))
