Activation.softplus constructor

Activation.softplus()

Softplus activation function

softplus(x) = ln(e^x+1)

Softplus is an antiderivative of sigmoid

Example:

final softplus = Activation.softplus();
final x = Matrix.row([-1, 0, 1]);
final y = softplus.function(x);
print(y); // output: matrix 1⨯3 [[0.31326168751822286, 0.6931471805599453, 1.3132616875182228]]

Implementation

Activation.softplus() {
  function = (Matrix m, [dynamic param]) =>
      m.apply((double x) => math.log(math.exp(x) + 1));
  dfunction = (Matrix m, [dynamic param]) =>
      [m.apply((double x) => 1 / (1 + math.exp(-x)))];
  name = 'softplus';
}