A Fortran-Keras Deep Learning Bridge for Scientific Computing

Algorithm 4

Implementation of crossentropy loss function and the corresponding derivation with respect to the input logits.

 real(rk) function crossentropy_loss(self, y_true, y_pred)

  ! Given predicted and expected output, returns the scalar loss

  class(network_type), intent(in out): self

  real(rk), intent(in): y_true(), y_pred()

  loss = - sum(y_true ∗ log(y_pred))

 end function loss

 function d_crossentropy_loss(self, y_true, y_pred) result(loss)

  ! Given predicted and expected output

  ! returns the loss with respect to softmax input

  class(network_type), intent(in out): self

  real(rk), intent(in): y_true(), y_pred()

  real(rk), allocatable: loss()

  loss = y_pred - y_true

 end function d_loss