Trait ort::tensor::ndarray_tensor::NdArrayExtensions

pub trait NdArrayExtensions<S, T, D> {
    // Required method
    fn softmax(&self, axis: Axis) -> Array<T, D>
       where D: RemoveAxis,
             S: RawData + Data + RawData<Elem = T>,
             <S as RawData>::Elem: Clone,
             T: NdFloat + SubAssign + DivAssign;
}

Trait extending ndarray::ArrayBase with useful tensor operations.

Generic

The trait is generic over:

• S: ndarray::ArrayBase’s data container
• T: Type contained inside the tensor (for example f32)
• D: Tensor’s dimension (ndarray::Dimension)

Required Methods

fn softmax(&self, axis: Axis) -> Array<T, D>where D: RemoveAxis, S: RawData + Data + RawData<Elem = T>, <S as RawData>::Elem: Clone, T: NdFloat + SubAssign + DivAssign,

Calculate the softmax of the tensor along a given axis

Trait Bounds

The function is generic and thus has some trait bounds:

• D: ndarray::RemoveAxis: The summation over an axis reduces the dimension of the tensor. A 0-D tensor thus cannot have a softmax calculated.
• S: ndarray::RawData + ndarray::Data + ndarray::RawData<Elem = T>: The storage of the tensor can be an owned array (ndarray::Array) or an array view (ndarray::ArrayView).
• <S as ndarray::RawData>::Elem: std::clone::Clone: The elements of the tensor must be Clone.
• T: ndarray::NdFloat + std::ops::SubAssign + std::ops::DivAssign: The elements of the tensor must be workable as floats and must support -= and /= operations.

Implementations on Foreign Types

impl<S, T, D> NdArrayExtensions<S, T, D> for ArrayBase<S, D>where D: RemoveAxis, S: RawData + Data + RawData<Elem = T>, <S as RawData>::Elem: Clone, T: NdFloat + SubAssign + DivAssign,

fn softmax(&self, axis: Axis) -> Array<T, D>

Implementors