Function autograd::ops::gradients [−] [src]

pub fn gradients(
    objective: &Tensor, 
    variables: &[&Tensor], 
    initial_grad: Option<&Tensor>
) -> Vec<Tensor>

Returns gradient tensors wrt variables.

Arguments

  • objective - Target of differentiation.
  • variables - Variable tensors with which differentiate objective.
  • initial_grad - This is required "if objective is not a scalar". In most cases, this is initialized with 1s.

Returns

Symbolic gradient tensors corresponding to variables in the same order as variables