如何在使用 ndarray 结构时正确满足 Rust 中的特征边界

Question

我正在尝试实现一个 Tensor 结构，它将保存 ndarray 箱中的数组，其中

T

是元素类型，

I

表示维度。该结构体的核心思想是保存一个二维或三维数组。

use std::fmt::Debug;
use ndarray::ArrayBase;
use ndarray::prelude::*;
use ndarray::OwnedRepr;
use ndarray::{Array};
use ndarray::Array3;
use ndarray::Dimension;

struct Tensor<T,I> 
where T: Clone + Sync + Send,
I: PartialEq + Debug 
{
   data: ArrayBase<OwnedRepr<T>,Dim<I>>
}


impl<T,I> Tensor<T,I>
where T: Clone + Sync + Send,
I: PartialEq + Debug 
{    
     pub fn new(arr:ArrayBase<OwnedRepr<T>,Dim<I>>) -> Self {
       Self {data: arr }
     }

     pub fn shape(&self) -> &[usize] {
       self.data.shape()
     }
}


fn main() {
    let mut temperature = Array3::<f32>::zeros((3, 4, 5));
    let shape = temperature.shape();
    let t3D:Tensor<f32,[usize;3]> = Tensor::new(temperature);
}

我想创建一个与

shape()

结构关联的

Tensor

方法，它返回驻留在

data

字段中的 ndarray 的形状。

但是

rustc

会抛出一个错误：

error[E0599]: the method `shape` exists for struct `ArrayBase<OwnedRepr<T>, Dim<I>>`, but its
trait bounds were not satisfied

**error[E0599]: the method `shape` exists for struct `ArrayBase<OwnedRepr<T>, Dim<I>>`, 
but its     trait bounds were not satisfied
--> src/main.rs:36:20
   |
36 |          self.data.shape()
   |                    ^^^^^ method cannot be called on `ArrayBase<OwnedRepr<T>, Dim<I>>` due to unsatisfied trait bounds**


doesn't satisfy `<_ as DimAdd<Dim<IxDynImpl>>>::Output = Dim<IxDynImpl>`
| doesn't satisfy `<_ as DimAdd<Dim<[usize; 0]>>>::Output = Dim<I>`
| doesn't satisfy `<_ as DimAdd<Dim<[usize; 1]>>>::Output = <Dim<I> as Dimension>::Larger`
| doesn't satisfy `<_ as DimMax<<Dim<I> as Dimension>::Larger>>::Output = <Dim<I> as Dimension>::Larger`
| doesn't satisfy `<_ as DimMax<<Dim<I> as Dimension>::Smaller>>::Output = Dim<I>`
| doesn't satisfy `<_ as DimMax<Dim<I>>>::Output = Dim<I>`
| doesn't satisfy `<_ as DimMax<Dim<IxDynImpl>>>::Output = Dim<IxDynImpl>`
| doesn't satisfy `<_ as DimMax<Dim<[usize; 0]>>>::Output = Dim<I>`
| doesn't satisfy `<_ as Index<usize>>::Output = usize`
| doesn't satisfy `<_ as Mul<usize>>::Output = Dim<I>`
| doesn't satisfy `<ndarray::Dim<I> as Add>::Output = ndarray::Dim<I>`
| doesn't satisfy `<ndarray::Dim<I> as Mul>::Output = ndarray::Dim<I>`
| doesn't satisfy `<ndarray::Dim<I> as Sub>::Output = ndarray::Dim<I>`
| doesn't satisfy `_: DimAdd<<Dim<I> as Dimension>::Larger>`
| doesn't satisfy `_: DimAdd<<Dim<I> as Dimension>::Smaller>`
| doesn't satisfy `_: DimMax<<Dim<I> as Dimension>::Larger>`
| doesn't satisfy `_: DimMax<<Dim<I> as Dimension>::Smaller>`
| doesn't satisfy `ndarray::Dim<I>: AddAssign<&'x ndarray::Dim<I>>`
| doesn't satisfy `ndarray::Dim<I>: AddAssign`
| doesn't satisfy `ndarray::Dim<I>: Add`
| doesn't satisfy `ndarray::Dim<I>: Clone`
| doesn't satisfy `ndarray::Dim<I>: Default`
| doesn't satisfy `ndarray::Dim<I>: DimAdd<ndarray::Dim<I>>`
| doesn't satisfy `ndarray::Dim<I>: DimAdd<ndarray::Dim<IxDynImpl>>`
| doesn't satisfy `ndarray::Dim<I>: DimAdd<ndarray::Dim<[usize; 0]>>`
| doesn't satisfy `ndarray::Dim<I>: DimAdd<ndarray::Dim<[usize; 1]>>`
| doesn't satisfy `ndarray::Dim<I>: DimMax<ndarray::Dim<I>>`
| doesn't satisfy `ndarray::Dim<I>: DimMax<ndarray::Dim<IxDynImpl>>`
| doesn't satisfy `ndarray::Dim<I>: DimMax<ndarray::Dim<[usize; 0]>>`
| doesn't satisfy `ndarray::Dim<I>: Dimension`
| doesn't satisfy `ndarray::Dim<I>: Eq`
| doesn't satisfy `ndarray::Dim<I>: IndexMut<usize>`
| doesn't satisfy `ndarray::Dim<I>: Mul<usize>`
| doesn't satisfy `ndarray::Dim<I>: MulAssign<&'x ndarray::Dim<I>>`
| doesn't satisfy `ndarray::Dim<I>: MulAssign<usize>`
| doesn't satisfy `ndarray::Dim<I>: MulAssign`
| doesn't satisfy `ndarray::Dim<I>: Mul`
| doesn't satisfy `ndarray::Dim<I>: Send`
| doesn't satisfy `ndarray::Dim<I>: SubAssign<&'x ndarray::Dim<I>>`
| doesn't satisfy `ndarray::Dim<I>: SubAssign`
| doesn't satisfy `ndarray::Dim<I>: Sub`
| doesn't satisfy `ndarray::Dim<I>: Sync`
| doesn't satisfy `ndarray::Dim<I>: std::ops::Index<usize>`

我尝试将特征边界应用于

struct

以及

impl

但编译器不断抛出错误

所以我有几个问题：

1. 如何将特征边界正确添加到上面的代码中？
2. 如何识别特定方法所需的特征绑定？

Answer 1

你可以约束

where Dim<I>: Dimension

。

Answer 2

最简单的方法是在

Dim<I>: Dimension

块上添加

impl

的边界（顺便说一句，请注意，您可以删除

struct

本身的边界）：

use ndarray::prelude::*;
use ndarray::Array;
use ndarray::Array3;
use ndarray::ArrayBase;
use ndarray::Dimension;
use ndarray::OwnedRepr;
use std::fmt::Debug;

struct Tensor<T, I> {
    data: ArrayBase<OwnedRepr<T>, Dim<I>>,
}

impl<T, I> Tensor<T, I>
where
    T: Clone + Sync + Send,
    I: PartialEq + Debug,
    Dim<I>: Dimension,
{
    pub fn new(arr: ArrayBase<OwnedRepr<T>, Dim<I>>) -> Self {
        Self { data: arr }
    }

    pub fn shape(&self) -> &[usize] {
        self.data.shape()
    }
}

fn main() {
    let mut temperature = Array3::<f32>::zeros((3, 4, 5));
    let shape = temperature.shape();
    let t3D: Tensor<f32, [usize; 3]> = Tensor::new(temperature);
}

游乐场

这可以从文档中推断出来，因为定义 impl ArrayBase

 方法

的 shape 块是：

impl<S, A, D> ArrayBase<S, D> where
    S: DataOwned<Elem = MaybeUninit<A>>,
    D: Dimension, 

在您的情况下，

D

是

Dim<I>

，所以您想要

Dim<I>: Dimension

。