Trait dfdx::tensor_ops::MaxTo

pub trait MaxTo: HasErr + HasShape {
    // Required method
    fn try_max<Dst: Shape, Ax: Axes>(
        self
    ) -> Result<Self::WithShape<Dst>, Self::Err>
       where Self::Shape: ReduceShapeTo<Dst, Ax>;

    // Provided method
    fn max<Dst: Shape, Ax: Axes>(self) -> Self::WithShape<Dst>
       where Self::Shape: ReduceShapeTo<Dst, Ax> { ... }
}

Reduction along multiple axes using max.

Required Methods

fn try_max<Dst: Shape, Ax: Axes>( self ) -> Result<Self::WithShape<Dst>, Self::Err>where Self::Shape: ReduceShapeTo<Dst, Ax>,

Fallible version of MaxTo::max

Provided Methods

fn max<Dst: Shape, Ax: Axes>(self) -> Self::WithShape<Dst>where Self::Shape: ReduceShapeTo<Dst, Ax>,

Max reduction. Pytorch equivalent: t.amax(Ax)

NOTE This evenly distributes gradients between all equal maximum values, instead of only exactly 1 value.

Example reducing a single axis:

let t: Tensor<Rank2<2, 3>, f32, _> = dev.tensor([[1.0, 2.0, 3.0], [-1.0, -2.0, -3.0]]);
let r = t.max::<Rank1<2>, _>(); // or `max::<_, Axis<1>>()`
assert_eq!(r.array(), [3.0, -1.0]);

Reducing multiple axes:

let r = t.max::<Rank0, _>();
assert_eq!(r.array(), 3.0);

Implementors

impl<S: Shape, E: Dtype, D: MaxReduceKernel<E>, T: Tape<E, D>> MaxTo for Tensor<S, E, D, T>