Create a one-dimensional array from a vector (no copying needed).

Panics if the length is greater than isize::MAX.

use ndarray::Array;

let array = Array::from_vec(vec![1., 2., 3., 4.]);

Create a one-dimensional array from an iterator or iterable.

Panics if the length is greater than isize::MAX.

use ndarray::Array;

let array = Array::from_iter(0..10);

Create a one-dimensional array with n evenly spaced elements from start to end (inclusive). A must be a floating point type.

Note that if start > end, the first element will still be start, and the following elements will be decreasing. This is different from the behavior of std::ops::RangeInclusive, which interprets start > end to mean that the range is empty.

Panics if n is greater than isize::MAX or if converting n - 1 to type A fails.

use ndarray::{Array, arr1};

let array = Array::linspace(0., 1., 5);
assert!(array == arr1(&[0.0, 0.25, 0.5, 0.75, 1.0]))

Create a one-dimensional array with elements from start to end (exclusive), incrementing by step. A must be a floating point type.

Panics if the length is greater than isize::MAX.

use ndarray::{Array, arr1};

let array = Array::range(0., 5., 1.);
assert!(array == arr1(&[0., 1., 2., 3., 4.]))

Create a one-dimensional array with n logarithmically spaced elements, with the starting value being base.powf(start) and the final one being base.powf(end). A must be a floating point type.

If base is negative, all values will be negative.

Panics if n is greater than isize::MAX or if converting n - 1 to type A fails.

use approx::assert_abs_diff_eq;
use ndarray::{Array, arr1};

let array = Array::logspace(10.0, 0.0, 3.0, 4);
assert_abs_diff_eq!(array, arr1(&[1e0, 1e1, 1e2, 1e3]), epsilon = 1e-12);

let array = Array::logspace(-10.0, 3.0, 0.0, 4);
assert_abs_diff_eq!(array, arr1(&[-1e3, -1e2, -1e1, -1e0]), epsilon = 1e-12);

Create a one-dimensional array with n geometrically spaced elements from start to end (inclusive). A must be a floating point type.

Returns None if start and end have different signs or if either one is zero. Conceptually, this means that in order to obtain a Some result, end / start must be positive.

Panics if n is greater than isize::MAX or if converting n - 1 to type A fails.

use approx::assert_abs_diff_eq;
use ndarray::{Array, arr1};

let array = Array::geomspace(1e0, 1e3, 4)?;
assert_abs_diff_eq!(array, arr1(&[1e0, 1e1, 1e2, 1e3]), epsilon = 1e-11);

let array = Array::geomspace(-1e3, -1e0, 4)?;
assert_abs_diff_eq!(array, arr1(&[-1e3, -1e2, -1e1, -1e0]), epsilon = 1e-11);

Create an identity matrix of size n (square 2D array).

Panics if n * n would overflow isize.

Create a 2D matrix from its diagonal

Panics if diag.len() * diag.len() would overflow isize.

use ndarray::{Array2, arr1, arr2};

let diag = arr1(&[1, 2]);
let array = Array2::from_diag(&diag);
assert_eq!(array, arr2(&[[1, 0], [0, 2]]));

Create a square 2D matrix of the specified size, with the specified element along the diagonal and zeros elsewhere.

Panics if n * n would overflow isize.

use ndarray::{array, Array2};

let array = Array2::from_diag_elem(2, 5.);
assert_eq!(array, array![[5., 0.], [0., 5.]]);

Create an array with copies of elem, shape shape.

Panics if the product of non-zero axis lengths overflows isize.

use ndarray::{Array, arr3, ShapeBuilder};

let a = Array::from_elem((2, 2, 2), 1.);

assert!(
    a == arr3(&[[[1., 1.],
                 [1., 1.]],
                [[1., 1.],
                 [1., 1.]]])
);
assert!(a.strides() == &[4, 2, 1]);

let b = Array::from_elem((2, 2, 2).f(), 1.);
assert!(b.strides() == &[1, 2, 4]);

Create an array with zeros, shape shape.

The element type is inferred; to control it, you can either specify type of the returned array or use turbofish syntax in the function call:

use ndarray::{Array1, arr1};

// Specify f32
let arr_f32: Array1<f32> = Array1::zeros(3);
assert_eq!(arr_f32, arr1(&[0_f32, 0.0, 0.0]));

// Specify i64
let arr_i64 = Array1::<i64>::zeros(3);
assert_eq!(arr_i64, arr1(&[0_i64, 0, 0]));

Panics if the product of non-zero axis lengths overflows isize.

Create an array with ones, shape shape.

Panics if the product of non-zero axis lengths overflows isize.

Create an array with default values, shape shape

Panics if the product of non-zero axis lengths overflows isize.

Create an array with values created by the function f.

f is called with no argument, and it should return the element to create. If the precise index of the element to create is needed, use from_shape_fn instead.

This constructor can be useful if the element order is not important, for example if they are identical or random.

Panics if the product of non-zero axis lengths overflows isize.

Create an array with values created by the function f.

f is called with the index of the element to create; the elements are visited in arbitrary order.

Panics if the product of non-zero axis lengths overflows isize.

use ndarray::{Array, arr2};

// Create a table of i × j (with i and j from 1 to 3)
let ij_table = Array::from_shape_fn((3, 3), |(i, j)| (1 + i) * (1 + j));

assert_eq!(
    ij_table,
    arr2(&[[1, 2, 3],
           [2, 4, 6],
           [3, 6, 9]])
);

Create an array with the given shape from a vector. (No cloning of elements needed.)

For a contiguous c- or f-order shape, the following applies:

Errors if shape does not correspond to the number of elements in v or if the shape/strides would result in overflowing isize.

For custom strides, the following applies:

Errors if strides and dimensions can point out of bounds of v, if strides allow multiple indices to point to the same element, or if the shape/strides would result in overflowing isize.

use ndarray::Array;
use ndarray::ShapeBuilder; // Needed for .strides() method
use ndarray::arr2;

let a = Array::from_shape_vec((2, 2), vec![1., 2., 3., 4.]);
assert!(a.is_ok());

let b = Array::from_shape_vec((2, 2).strides((1, 2)),
                              vec![1., 2., 3., 4.]).unwrap();
assert!(
    b == arr2(&[[1., 3.],
                [2., 4.]])
);

Creates an array from a vector and interpret it according to the provided shape and strides. (No cloning of elements needed.)

§Safety

The caller must ensure that the following conditions are met:

1. The ndim of dim and strides must be the same.

2. The product of non-zero axis lengths must not exceed isize::MAX.

3. For axes with length > 1, the pointer cannot move outside the slice.

4. If the array will be empty (any axes are zero-length), the difference between the least address and greatest address accessible by moving along all axes must be ≤ v.len().

   If the array will not be empty, the difference between the least address and greatest address accessible by moving along all axes must be < v.len().

5. The strides must not allow any element to be referenced by two different indices.

Create an array with uninitialized elements, shape shape.

The uninitialized elements of type A are represented by the type MaybeUninit<A>, an easier way to handle uninit values correctly.

Only when the array is completely initialized with valid elements, can it be converted to an array of A elements using .assume_init().

Panics if the number of elements in shape would overflow isize.

§Safety

The whole of the array must be initialized before it is converted using .assume_init() or otherwise traversed/read with the element type A.

§Examples

It is possible to assign individual values through *elt = MaybeUninit::new(value) and so on.

use ndarray::{s, Array2};

// Example Task: Let's create a column shifted copy of the input

fn shift_by_two(a: &Array2<f32>) -> Array2<f32> {
    // create an uninitialized array
    let mut b = Array2::uninit(a.dim());

    // two first columns in b are two last in a
    // rest of columns in b are the initial columns in a

    a.slice(s![.., -2..]).assign_to(b.slice_mut(s![.., ..2]));
    a.slice(s![.., 2..]).assign_to(b.slice_mut(s![.., ..-2]));

    // Now we can promise that `b` is safe to use with all operations
    unsafe {
        b.assume_init()
    }
}

Create an array with uninitialized elements, shape shape.

The uninitialized elements of type A are represented by the type MaybeUninit<A>, an easier way to handle uninit values correctly.

The builder closure gets unshared access to the array through a view and can use it to modify the array before it is returned. This allows initializing the array for any owned array type (avoiding clone requirements for copy-on-write, because the array is unshared when initially created).

Only when the array is completely initialized with valid elements, can it be converted to an array of A elements using .assume_init().

Panics if the number of elements in shape would overflow isize.

§Safety

The whole of the array must be initialized before it is converted using .assume_init() or otherwise traversed/read with the element type A.

Return an uniquely owned copy of the array.

If the input array is contiguous, then the output array will have the same memory layout. Otherwise, the layout of the output array is unspecified. If you need a particular layout, you can allocate a new array with the desired memory layout and .assign() the data. Alternatively, you can collectan iterator, like this for a result in standard layout:

Array::from_shape_vec(arr.raw_dim(), arr.iter().cloned().collect()).unwrap()

or this for a result in column-major (Fortran) layout:

Array::from_shape_vec(arr.raw_dim().f(), arr.t().iter().cloned().collect()).unwrap()

Converts the array into Array<A, D> if this is possible without cloning the array elements. Otherwise, returns self unchanged.

use ndarray::{array, rcarr2, ArcArray2, Array2};

// Reference-counted, clone-on-write `ArcArray`.
let a: ArcArray2<_> = rcarr2(&[[1., 2.], [3., 4.]]);
{
    // Another reference to the same data.
    let b: ArcArray2<_> = a.clone();
    // Since there are two references to the same data, `.into_owned()`
    // would require cloning the data, so `.try_into_owned_nocopy()`
    // returns `Err`.
    assert!(b.try_into_owned_nocopy().is_err());
}
// Here, since the second reference has been dropped, the `ArcArray`
// can be converted into an `Array` without cloning the data.
let unique: Array2<_> = a.try_into_owned_nocopy().unwrap();
assert_eq!(unique, array![[1., 2.], [3., 4.]]);

Turn the array into a shared ownership (copy on write) array, cloning the array elements if necessary.

If you want to generalize over Array and ArcArray inputs but avoid an A: Clone bound, use Into::<ArcArray<A, D>>::into instead of this method.

Slice the array, possibly changing the number of dimensions.

See Slicing for full documentation. See also s!, SliceArg, and SliceInfo.

Panics if an index is out of bounds or step size is zero.
(Panics if D is IxDyn and info does not match the number of array axes.)

Slice the array in place along the specified axis, then return the sliced array.

Panics if an index is out of bounds or step size is zero.
Panics if axis is out of bounds.

Collapses the array to index along the axis and removes the axis.

See .index_axis() and Subviews for full documentation.

Panics if axis or index is out of bounds.

Return a mutable pointer to the first element in the array.

This method attempts to unshare the data. If S: DataMut, then the data is guaranteed to be uniquely held on return.

§Warning

When accessing elements through this pointer, make sure to use strides obtained after calling this method, since the process of unsharing the data may change the strides.

Return a raw mutable view of the array.

This method attempts to unshare the data. If S: DataMut, then the data is guaranteed to be uniquely held on return.

Return the array’s data as a slice if it is contiguous, return None otherwise.

In the contiguous case, in order to return a unique reference, this method unshares the data if necessary, but it preserves the existing strides.

Transform the array into shape; any shape with the same number of elements is accepted, but the source array must be contiguous.

If an index ordering is not specified, the default is RowMajor. The operation will only succeed if the array’s memory layout is compatible with the index ordering, so that the array elements can be rearranged in place.

If required use .to_shape() or .into_shape_clone instead for more flexible reshaping of arrays, which allows copying elements if required.

Errors if the shapes don’t have the same number of elements.
Errors if order RowMajor is given but input is not c-contiguous. Errors if order ColumnMajor is given but input is not f-contiguous.

If shape is not given: use memory layout of incoming array. Row major arrays are reshaped using row major index ordering, column major arrays with column major index ordering.

The new_shape parameter should be a dimension and an optional order like these examples:

(3, 4)                          // Shape 3 x 4 with default order (RowMajor)
((3, 4), Order::RowMajor))      // use specific order
((3, 4), Order::ColumnMajor))   // use specific order
((3, 4), Order::C))             // use shorthand for order - shorthands C and F

§Example

use ndarray::{aview1, aview2};
use ndarray::Order;

assert!(
    aview1(&[1., 2., 3., 4.]).into_shape_with_order((2, 2)).unwrap()
    == aview2(&[[1., 2.],
                [3., 4.]])
);

assert!(
    aview1(&[1., 2., 3., 4.]).into_shape_with_order(((2, 2), Order::ColumnMajor)).unwrap()
    == aview2(&[[1., 3.],
                [2., 4.]])
);

Transform the array into shape; any shape with the same number of elements is accepted, but the source array or view must be in standard or column-major (Fortran) layout.

Note that .into_shape() “moves” elements differently depending on if the input array is C-contig or F-contig, it follows the index order that corresponds to the memory order. Prefer to use .to_shape() or .into_shape_with_order().

Because of this, the method is deprecated. That reshapes depend on memory order is not intuitive.

Errors if the shapes don’t have the same number of elements.
Errors if the input array is not c- or f-contiguous.

use ndarray::{aview1, aview2};

assert!(
    aview1(&[1., 2., 3., 4.]).into_shape((2, 2)).unwrap()
    == aview2(&[[1., 2.],
                [3., 4.]])
);

Transform the array into shape; any shape with the same number of elements is accepted. Array elements are reordered in place if possible, otherwise they are copied to create a new array.

If an index ordering is not specified, the default is RowMajor.

§.to_shape vs .into_shape_clone

• to_shape supports views and outputting views
• to_shape borrows the original array, into_shape_clone consumes the original
• into_shape_clone preserves array type (Array vs ArcArray), but does not support views.

Errors if the shapes don’t have the same number of elements.

Note: Reshape is for ArcArray only. Use .into_shape_with_order() for other arrays and array views.

Transform the array into shape; any shape with the same number of elements is accepted.

May clone all elements if needed to arrange elements in standard layout (and break sharing).

Panics if shapes are incompatible.

This method is obsolete, because it is inflexible in how logical order of the array is handled. See ArrayRef::to_shape().

use ndarray::{rcarr1, rcarr2};

assert!(
    rcarr1(&[1., 2., 3., 4.]).reshape((2, 2))
    == rcarr2(&[[1., 2.],
                [3., 4.]])
);

Flatten the array to a one-dimensional array, consuming the array.

If possible, no copy is made, and the new array use the same memory as the original array. Otherwise, a new array is allocated and the elements are copied.

use ndarray::{arr1, arr3};

let array = arr3(&[[[1, 2], [3, 4]], [[5, 6], [7, 8]]]);
let flattened = array.into_flat();
assert_eq!(flattened, arr1(&[1, 2, 3, 4, 5, 6, 7, 8]));

Convert any array or array view to a dynamic dimensional array or array view (respectively).

use ndarray::{arr2, ArrayD};

let array: ArrayD<i32> = arr2(&[[1, 2],
                                [3, 4]]).into_dyn();

Convert an array or array view to another with the same type, but different dimensionality type. Errors if the dimensions don’t agree (the number of axes must match).

Note that conversion to a dynamic dimensional array will never fail (and is equivalent to the into_dyn method).

use ndarray::{ArrayD, Ix2, IxDyn};

// Create a dynamic dimensionality array and convert it to an Array2
// (Ix2 dimension type).

let array = ArrayD::<f64>::zeros(IxDyn(&[10, 10]));

assert!(array.into_dimensionality::<Ix2>().is_ok());

Permute the axes.

This does not move any data, it just adjusts the array’s dimensions and strides.

i in the j-th place in the axes sequence means self’s i-th axis becomes self.permuted_axes()’s j-th axis

Panics if any of the axes are out of bounds, if an axis is missing, or if an axis is repeated more than once.

§Examples

use ndarray::{arr2, Array3};

let a = arr2(&[[0, 1], [2, 3]]);
assert_eq!(a.view().permuted_axes([1, 0]), a.t());

let b = Array3::<u8>::zeros((1, 2, 3));
assert_eq!(b.permuted_axes([1, 0, 2]).shape(), &[2, 1, 3]);

Permute the axes in-place.

This does not move any data, it just adjusts the array’s dimensions and strides.

i in the j-th place in the axes sequence means self’s i-th axis becomes self’s j-th axis

Panics if any of the axes are out of bounds, if an axis is missing, or if an axis is repeated more than once.

§Example

use ndarray::{arr2, Array3};

let mut a = arr2(&[[0, 1], [2, 3]]);
a.permute_axes([1, 0]);
assert_eq!(a, arr2(&[[0, 2], [1, 3]]));

let mut b = Array3::<u8>::zeros((1, 2, 3));
b.permute_axes([1, 0, 2]);
assert_eq!(b.shape(), &[2, 1, 3]);

Transpose the array by reversing axes.

Transposition reverses the order of the axes (dimensions and strides) while retaining the same data.

Reverse the axes of the array in-place.

This does not move any data, it just adjusts the array’s dimensions and strides.

Insert new array axis at axis and return the result.

use ndarray::{Array3, Axis, arr1, arr2};

// Convert a 1-D array into a row vector (2-D).
let a = arr1(&[1, 2, 3]);
let row = a.insert_axis(Axis(0));
assert_eq!(row, arr2(&[[1, 2, 3]]));

// Convert a 1-D array into a column vector (2-D).
let b = arr1(&[1, 2, 3]);
let col = b.insert_axis(Axis(1));
assert_eq!(col, arr2(&[[1], [2], [3]]));

// The new axis always has length 1.
let b = Array3::<f64>::zeros((3, 4, 5));
assert_eq!(b.insert_axis(Axis(2)).shape(), &[3, 4, 1, 5]);

Panics if the axis is out of bounds.

Remove array axis axis and return the result.

This is equivalent to .index_axis_move(axis, 0) and makes most sense to use if the axis to remove is of length 1.

Panics if the axis is out of bounds or its length is zero.

Call f by value on each element, update the array with the new values and return it.

Elements are visited in arbitrary order.

Consume the array, call f by value on each element, and return an owned array with the new values. Works for any F: FnMut(A)->B.

If A and B are the same type then the map is performed by delegating to mapv_into and then converting into an owned array. This avoids unnecessary memory allocations in mapv.

If A and B are different types then a new array is allocated and the map is performed as in mapv.

Elements are visited in arbitrary order.

Return a raw pointer to the element at index, or return None if the index is out of bounds.

use ndarray::arr2;

let a = arr2(&[[1., 2.], [3., 4.]]);

let v = a.raw_view();
let p = a.get_ptr((0, 1)).unwrap();

assert_eq!(unsafe { *p }, 2.);

Return a raw pointer to the element at index, or return None if the index is out of bounds.

use ndarray::arr2;

let mut a = arr2(&[[1., 2.], [3., 4.]]);

let v = a.raw_view_mut();
let p = a.get_mut_ptr((0, 1)).unwrap();

unsafe {
    *p = 5.;
}

assert_eq!(a.get((0, 1)), Some(&5.));

Return a pointer to the first element in the array.

Raw access to array elements needs to follow the strided indexing scheme: an element at multi-index I in an array with strides S is located at offset

Σ_{0 ≤ k < d} I_k × S_k

where d is self.ndim().

Slice the array in place without changing the number of dimensions.

In particular, if an axis is sliced with an index, the axis is collapsed, as in .collapse_axis(), rather than removed, as in .slice_move() or .index_axis_move().

See Slicing for full documentation. See also s!, SliceArg, and SliceInfo.

Panics in the following cases:

• if an index is out of bounds
• if a step size is zero
• if NewAxis is in info, e.g. if NewAxis was used in the s! macro
• if D is IxDyn and info does not match the number of array axes

Slice the array in place along the specified axis.

Panics if an index is out of bounds or step size is zero.
Panics if axis is out of bounds.

Slice the array in place, with a closure specifying the slice for each axis.

This is especially useful for code which is generic over the dimensionality of the array.

Panics if an index is out of bounds or step size is zero.

Selects index along the axis, collapsing the axis into length one.

Panics if axis or index is out of bounds.

Return true if the array data is laid out in contiguous “C order” in memory (where the last index is the most rapidly varying).

Return false otherwise, i.e. the array is possibly not contiguous in memory, it has custom strides, etc.

Reverse the stride of axis.

Panics if the axis is out of bounds.

Swap axes ax and bx.

This does not move any data, it just adjusts the array’s dimensions and strides.

Panics if the axes are out of bounds.

use ndarray::arr2;

let mut a = arr2(&[[1., 2., 3.]]);
a.swap_axes(0, 1);
assert!(
    a == arr2(&[[1.], [2.], [3.]])
);

If possible, merge in the axis take to into.

Returns true iff the axes are now merged.

This method merges the axes if movement along the two original axes (moving fastest along the into axis) can be equivalently represented as movement along one (merged) axis. Merging the axes preserves this order in the merged axis. If take and into are the same axis, then the axis is “merged” if its length is ≤ 1.

If the return value is true, then the following hold:

• The new length of the into axis is the product of the original lengths of the two axes.

• The new length of the take axis is 0 if the product of the original lengths of the two axes is 0, and 1 otherwise.

If the return value is false, then merging is not possible, and the original shape and strides have been preserved.

Note that the ordering constraint means that if it’s possible to merge take into into, it’s usually not possible to merge into into take, and vice versa.

use ndarray::Array3;
use ndarray::Axis;

let mut a = Array3::<f64>::zeros((2, 3, 4));
assert!(a.merge_axes(Axis(1), Axis(2)));
assert_eq!(a.shape(), &[2, 1, 12]);

Panics if an axis is out of bounds.

Return the length of axis.

The axis should be in the range Axis( 0 .. n ) where n is the number of dimensions (axes) of the array.

Panics if the axis is out of bounds.

Return the shape of the array as it’s stored in the array.

This is primarily useful for passing to other ArrayBase functions, such as when creating another array of the same shape and dimensionality.

use ndarray::Array;

let a = Array::from_elem((2, 3), 5.);

// Create an array of zeros that's the same shape and dimensionality as `a`.
let b = Array::<f64, _>::zeros(a.raw_dim());

Return the shape of the array as a slice.

Note that you probably don’t want to use this to create an array of the same shape as another array because creating an array with e.g. Array::zeros() using a shape of type &[usize] results in a dynamic-dimensional array. If you want to create an array that has the same shape and dimensionality as another array, use .raw_dim() instead:

use ndarray::{Array, Array2};

let a = Array2::<i32>::zeros((3, 4));
let shape = a.shape();
assert_eq!(shape, &[3, 4]);

// Since `a.shape()` returned `&[usize]`, we get an `ArrayD` instance:
let b = Array::zeros(shape);
assert_eq!(a.clone().into_dyn(), b);

// To get the same dimension type, use `.raw_dim()` instead:
let c = Array::zeros(a.raw_dim());
assert_eq!(a, c);

Return the stride of axis.

The axis should be in the range Axis( 0 .. n ) where n is the number of dimensions (axes) of the array.

Panics if the axis is out of bounds.

Returns the single element in the array without cloning it.

use ndarray::{arr0, Array0};

// `Foo` doesn't implement `Clone`.
#[derive(Debug, Eq, PartialEq)]
struct Foo;

let array: Array0<Foo> = arr0(Foo);
let scalar: Foo = array.into_scalar();
assert_eq!(scalar, Foo);

Return a vector of the elements in the array, in the way they are stored internally, and the index in the vector corresponding to the logically first element of the array (or None if the array is empty).

If the array is in standard memory layout, the logical element order of the array (.iter() order) and of the returned vector will be the same.

use ndarray::{array, Array2, Axis};

let mut arr: Array2<f64> = array![[1., 2.], [3., 4.], [5., 6.]];
arr.slice_axis_inplace(Axis(0), (1..).into());
assert_eq!(arr[[0, 0]], 3.);
let copy = arr.clone();

let shape = arr.shape().to_owned();
let strides = arr.strides().to_owned();
let (v, offset) = arr.into_raw_vec_and_offset();

assert_eq!(v, &[1., 2., 3., 4., 5., 6.]);
assert_eq!(offset, Some(2));
assert_eq!(v[offset.unwrap()], 3.);
for row in 0..shape[0] {
    for col in 0..shape[1] {
        let index = (
            offset.unwrap() as isize
            + row as isize * strides[0]
            + col as isize * strides[1]
        ) as usize;
        assert_eq!(v[index], copy[[row, col]]);
    }
}

In the case of zero-sized elements, the offset to the logically first element is somewhat meaningless. For convenience, an offset will be returned such that all indices computed using the offset, shape, and strides will be in-bounds for the Vec<A>. Note that this offset won’t necessarily be the same as the offset for an array of nonzero-sized elements sliced in the same way.

use ndarray::{array, Array2, Axis};

let mut arr: Array2<()> = array![[(), ()], [(), ()], [(), ()]];
arr.slice_axis_inplace(Axis(0), (1..).into());

let shape = arr.shape().to_owned();
let strides = arr.strides().to_owned();
let (v, offset) = arr.into_raw_vec_and_offset();

assert_eq!(v, &[(), (), (), (), (), ()]);
for row in 0..shape[0] {
    for col in 0..shape[1] {
        let index = (
            offset.unwrap() as isize
            + row as isize * strides[0]
            + col as isize * strides[1]
        ) as usize;
        assert_eq!(v[index], ());
    }
}

Return a vector of the elements in the array, in the way they are stored internally.

Depending on slicing and strides, the logically first element of the array can be located at an offset. Because of this, prefer to use .into_raw_vec_and_offset() instead.

Append a row to an array

The elements from row are cloned and added as a new row in the array.

Errors with a shape error if the length of the row does not match the length of the rows in the array.

The memory layout of the self array matters for ensuring that the append is efficient. Appending automatically changes memory layout of the array so that it is appended to along the “growing axis”. However, if the memory layout needs adjusting, the array must reallocate and move memory.

The operation leaves the existing data in place and is most efficient if one of these is true:

• The axis being appended to is the longest stride axis, i.e the array is in row major (“C”) layout.
• The array has 0 or 1 rows (It is converted to row major)

Ensure appending is efficient by, for example, appending to an empty array and then always pushing/appending along the same axis. For pushing rows, ndarray’s default layout (C order) is efficient.

When repeatedly appending to a single axis, the amortized average complexity of each append is O(m), where m is the length of the row.

use ndarray::{Array, ArrayView, array};

// create an empty array and append
let mut a = Array::zeros((0, 4));
a.push_row(ArrayView::from(&[ 1.,  2.,  3.,  4.])).unwrap();
a.push_row(ArrayView::from(&[-1., -2., -3., -4.])).unwrap();

assert_eq!(
    a,
    array![[ 1.,  2.,  3.,  4.],
           [-1., -2., -3., -4.]]);

Append a column to an array

The elements from column are cloned and added as a new column in the array.

Errors with a shape error if the length of the column does not match the length of the columns in the array.

The memory layout of the self array matters for ensuring that the append is efficient. Appending automatically changes memory layout of the array so that it is appended to along the “growing axis”. However, if the memory layout needs adjusting, the array must reallocate and move memory.

The operation leaves the existing data in place and is most efficient if one of these is true:

• The axis being appended to is the longest stride axis, i.e the array is in column major (“F”) layout.
• The array has 0 or 1 columns (It is converted to column major)

Ensure appending is efficient by, for example, appending to an empty array and then always pushing/appending along the same axis. For pushing columns, column major layout (F order) is efficient.

When repeatedly appending to a single axis, the amortized average complexity of each append is O(m), where m is the length of the column.

use ndarray::{Array, ArrayView, array};

// create an empty array and append
let mut a = Array::zeros((2, 0));
a.push_column(ArrayView::from(&[1., 2.])).unwrap();
a.push_column(ArrayView::from(&[-1., -2.])).unwrap();

assert_eq!(
    a,
    array![[1., -1.],
           [2., -2.]]);

Reserve capacity to grow array by at least additional rows.

Existing elements of array are untouched and the backing storage is grown by calling the underlying reserve method of the OwnedRepr.

This is useful when pushing or appending repeatedly to an array to avoid multiple allocations.

Errors with a shape error if the resultant capacity is larger than the addressable bounds; that is, the product of non-zero axis lengths once axis has been extended by additional exceeds isize::MAX.

use ndarray::Array2;
let mut a = Array2::<i32>::zeros((2,4));
a.reserve_rows(1000).unwrap();
assert!(a.into_raw_vec().capacity() >= 4*1002);

Reserve capacity to grow array by at least additional columns.

Existing elements of array are untouched and the backing storage is grown by calling the underlying reserve method of the OwnedRepr.

This is useful when pushing or appending repeatedly to an array to avoid multiple allocations.

Errors with a shape error if the resultant capacity is larger than the addressable bounds; that is, the product of non-zero axis lengths once axis has been extended by additional exceeds isize::MAX.

use ndarray::Array2;
let mut a = Array2::<i32>::zeros((2,4));
a.reserve_columns(1000).unwrap();
assert!(a.into_raw_vec().capacity() >= 2*1002);

Move all elements from self into new_array, which must be of the same shape but can have a different memory layout. The destination is overwritten completely.

The destination should be a mut reference to an array or an ArrayViewMut with A elements.

Panics if the shapes don’t agree.

§Example

use ndarray::Array;

// Usage example of move_into in safe code
let mut a = Array::default((10, 10));
let b = Array::from_shape_fn((10, 10), |(i, j)| (i + j).to_string());
b.move_into(&mut a);

Move all elements from self into new_array, which must be of the same shape but can have a different memory layout. The destination is overwritten completely.

The destination should be a mut reference to an array or an ArrayViewMut with MaybeUninit<A> elements (which are overwritten without dropping any existing value).

Minor implementation note: Owned arrays like self may be sliced in place and own elements that are not part of their active view; these are dropped at the end of this function, after all elements in the “active view” are moved into new_array. If there is a panic in drop of any such element, other elements may be leaked.

Panics if the shapes don’t agree.

§Example

use ndarray::Array;

let a = Array::from_iter(0..100).into_shape_with_order((10, 10)).unwrap();
let mut b = Array::uninit((10, 10));
a.move_into_uninit(&mut b);
unsafe {
    // we can now promise we have fully initialized `b`.
    let b = b.assume_init();
}

Append an array to the array along an axis.

The elements of array are cloned and extend the axis axis in the present array; self will grow in size by 1 along axis.

Append to the array, where the array being pushed to the array has one dimension less than the self array. This method is equivalent to append in this way: self.append(axis, array.insert_axis(axis)).

Errors with a shape error if the shape of self does not match the array-to-append; all axes except the axis along which it being appended matter for this check: the shape of self with axis removed must be the same as the shape of array.

The memory layout of the self array matters for ensuring that the append is efficient. Appending automatically changes memory layout of the array so that it is appended to along the “growing axis”. However, if the memory layout needs adjusting, the array must reallocate and move memory.

The operation leaves the existing data in place and is most efficient if axis is a “growing axis” for the array, i.e. one of these is true:

• The axis is the longest stride axis, for example the 0th axis in a C-layout or the n-1th axis in an F-layout array.
• The axis has length 0 or 1 (It is converted to the new growing axis)

Ensure appending is efficient by for example starting from an empty array and/or always appending to an array along the same axis.

The amortized average complexity of the append, when appending along its growing axis, is O(m) where m is the number of individual elements to append.

The memory layout of the argument array does not matter to the same extent.

use ndarray::{Array, ArrayView, array, Axis};

// create an empty array and push rows to it
let mut a = Array::zeros((0, 4));
let ones  = ArrayView::from(&[1.; 4]);
let zeros = ArrayView::from(&[0.; 4]);
a.push(Axis(0), ones).unwrap();
a.push(Axis(0), zeros).unwrap();
a.push(Axis(0), ones).unwrap();

assert_eq!(
    a,
    array![[1., 1., 1., 1.],
           [0., 0., 0., 0.],
           [1., 1., 1., 1.]]);

Append an array to the array along an axis.

The elements of array are cloned and extend the axis axis in the present array; self will grow in size by array.len_of(axis) along axis.

Errors with a shape error if the shape of self does not match the array-to-append; all axes except the axis along which it being appended matter for this check: the shape of self with axis removed must be the same as the shape of array with axis removed.

The memory layout of the self array matters for ensuring that the append is efficient. Appending automatically changes memory layout of the array so that it is appended to along the “growing axis”. However, if the memory layout needs adjusting, the array must reallocate and move memory.

The operation leaves the existing data in place and is most efficient if axis is a “growing axis” for the array, i.e. one of these is true:

• The axis is the longest stride axis, for example the 0th axis in a C-layout or the n-1th axis in an F-layout array.
• The axis has length 0 or 1 (It is converted to the new growing axis)

Ensure appending is efficient by for example starting from an empty array and/or always appending to an array along the same axis.

The amortized average complexity of the append, when appending along its growing axis, is O(m) where m is the number of individual elements to append.

The memory layout of the argument array does not matter to the same extent.

use ndarray::{Array, ArrayView, array, Axis};

// create an empty array and append two rows at a time
let mut a = Array::zeros((0, 4));
let ones  = ArrayView::from(&[1.; 8]).into_shape_with_order((2, 4)).unwrap();
let zeros = ArrayView::from(&[0.; 8]).into_shape_with_order((2, 4)).unwrap();
a.append(Axis(0), ones).unwrap();
a.append(Axis(0), zeros).unwrap();
a.append(Axis(0), ones).unwrap();

assert_eq!(
    a,
    array![[1., 1., 1., 1.],
           [1., 1., 1., 1.],
           [0., 0., 0., 0.],
           [0., 0., 0., 0.],
           [1., 1., 1., 1.],
           [1., 1., 1., 1.]]);

Reserve capacity to grow array along axis by at least additional elements.

The axis should be in the range Axis( 0 .. n ) where n is the number of dimensions (axes) of the array.

Existing elements of array are untouched and the backing storage is grown by calling the underlying reserve method of the OwnedRepr.

This is useful when pushing or appending repeatedly to an array to avoid multiple allocations.

Panics if the axis is out of bounds.

Errors with a shape error if the resultant capacity is larger than the addressable bounds; that is, the product of non-zero axis lengths once axis has been extended by additional exceeds isize::MAX.

use ndarray::{Array3, Axis};
let mut a = Array3::<i32>::zeros((0,2,4));
a.reserve(Axis(0), 1000).unwrap();
assert!(a.into_raw_vec().capacity() >= 2*4*1000);

Promise that the array’s elements are all fully initialized, and convert the array from element type MaybeUninit<A> to A.

For example, it can convert an Array<MaybeUninit<f64>, D> to Array<f64, D>.

§Safety

Safe to use if all the array’s elements have been initialized.

Note that for owned and shared ownership arrays, the promise must include all of the array’s storage; it is for example possible to slice these in place, but that must only be done after all elements have been initialized.

Return the number of rows (length of Axis(0)) in the two-dimensional array.

use ndarray::{array, Axis};

let array = array![[1., 2.],
                   [3., 4.],
                   [5., 6.]];
assert_eq!(array.nrows(), 3);

// equivalent ways of getting the dimensions
// get nrows, ncols by using dim:
let (m, n) = array.dim();
assert_eq!(m, array.nrows());
// get length of any particular axis with .len_of()
assert_eq!(m, array.len_of(Axis(0)));

Return the number of columns (length of Axis(1)) in the two-dimensional array.

use ndarray::{array, Axis};

let array = array![[1., 2.],
                   [3., 4.],
                   [5., 6.]];
assert_eq!(array.ncols(), 2);

// equivalent ways of getting the dimensions
// get nrows, ncols by using dim:
let (m, n) = array.dim();
assert_eq!(n, array.ncols());
// get length of any particular axis with .len_of()
assert_eq!(n, array.len_of(Axis(1)));

Return true if the array is square, false otherwise.

§Examples

Square:

use ndarray::array;
let array = array![[1., 2.], [3., 4.]];
assert!(array.is_square());

Not square:

use ndarray::array;
let array = array![[1., 2., 5.], [3., 4., 6.]];
assert!(!array.is_square());

Insert new array axis of length 1 at axis, modifying the shape and strides in-place.

Panics if the axis is out of bounds.

use ndarray::{Axis, arr2, arr3};

let mut a = arr2(&[[1, 2, 3], [4, 5, 6]]).into_dyn();
assert_eq!(a.shape(), &[2, 3]);

a.insert_axis_inplace(Axis(1));
assert_eq!(a, arr3(&[[[1, 2, 3]], [[4, 5, 6]]]).into_dyn());
assert_eq!(a.shape(), &[2, 1, 3]);

Collapses the array to index along the axis and removes the axis, modifying the shape and strides in-place.

Panics if axis or index is out of bounds.

use ndarray::{Axis, arr1, arr2};

let mut a = arr2(&[[1, 2, 3], [4, 5, 6]]).into_dyn();
assert_eq!(a.shape(), &[2, 3]);

a.index_axis_inplace(Axis(1), 1);
assert_eq!(a, arr1(&[2, 5]).into_dyn());
assert_eq!(a.shape(), &[2]);

Remove axes of length 1 and return the modified array.

If the array has more the one dimension, the result array will always have at least one dimension, even if it has a length of 1.

use ndarray::{arr1, arr2, arr3};

let a = arr3(&[[[1, 2, 3]], [[4, 5, 6]]]).into_dyn();
assert_eq!(a.shape(), &[2, 1, 3]);
let b = a.squeeze();
assert_eq!(b, arr2(&[[1, 2, 3], [4, 5, 6]]).into_dyn());
assert_eq!(b.shape(), &[2, 3]);

let c = arr2(&[[1]]).into_dyn();
assert_eq!(c.shape(), &[1, 1]);
let d = c.squeeze();
assert_eq!(d, arr1(&[1]).into_dyn());
assert_eq!(d.shape(), &[1]);

Create a read-only array view borrowing its data from a slice.

Checks whether shape are compatible with the slice’s length, returning an Err if not compatible.

use ndarray::ArrayView;
use ndarray::arr3;
use ndarray::ShapeBuilder;

// advanced example where we are even specifying exact strides to use (which is optional).
let s = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12];
let a = ArrayView::from_shape((2, 3, 2).strides((1, 4, 2)),
                              &s).unwrap();

assert!(
    a == arr3(&[[[0, 2],
                 [4, 6],
                 [8, 10]],
                [[1, 3],
                 [5, 7],
                 [9, 11]]])
);
assert!(a.strides() == &[1, 4, 2]);

Create an ArrayView<A, D> from shape information and a raw pointer to the elements.

§Safety

The caller is responsible for ensuring all of the following:

• The elements seen by moving ptr according to the shape and strides must live at least as long as 'a and must not be not mutably aliased for the duration of 'a.

• ptr must be non-null and aligned, and it must be safe to .offset() ptr by zero.

• It must be safe to .offset() the pointer repeatedly along all axes and calculate the counts for the .offset() calls without overflow, even if the array is empty or the elements are zero-sized.

  In other words,

  • All possible pointers generated by moving along all axes must be in bounds or one byte past the end of a single allocation with element type A. The only exceptions are if the array is empty or the element type is zero-sized. In these cases, ptr may be dangling, but it must still be safe to .offset() the pointer along the axes.

  • The offset in units of bytes between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents the computed offset, in bytes, from overflowing isize regardless of the starting point due to past offsets.

  • The offset in units of A between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents overflow when calculating the count parameter to .offset() regardless of the starting point due to past offsets.

• The product of non-zero axis lengths must not exceed isize::MAX.

• Strides must be non-negative.

This function can use debug assertions to check some of these requirements, but it’s not a complete check.

Create a read-write array view borrowing its data from a slice.

Checks whether dim and strides are compatible with the slice’s length, returning an Err if not compatible.

use ndarray::ArrayViewMut;
use ndarray::arr3;
use ndarray::ShapeBuilder;

let mut s = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12];
let mut a = ArrayViewMut::from_shape((2, 3, 2).strides((1, 4, 2)),
                                     &mut s).unwrap();

a[[0, 0, 0]] = 1;
assert!(
    a == arr3(&[[[1, 2],
                 [4, 6],
                 [8, 10]],
                [[1, 3],
                 [5, 7],
                 [9, 11]]])
);
assert!(a.strides() == &[1, 4, 2]);

Create an ArrayViewMut<A, D> from shape information and a raw pointer to the elements.

§Safety

The caller is responsible for ensuring all of the following:

• The elements seen by moving ptr according to the shape and strides must live at least as long as 'a and must not be aliased for the duration of 'a.

• ptr must be non-null and aligned, and it must be safe to .offset() ptr by zero.

• It must be safe to .offset() the pointer repeatedly along all axes and calculate the counts for the .offset() calls without overflow, even if the array is empty or the elements are zero-sized.

  In other words,

  • All possible pointers generated by moving along all axes must be in bounds or one byte past the end of a single allocation with element type A. The only exceptions are if the array is empty or the element type is zero-sized. In these cases, ptr may be dangling, but it must still be safe to .offset() the pointer along the axes.

  • The offset in units of bytes between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents the computed offset, in bytes, from overflowing isize regardless of the starting point due to past offsets.

  • The offset in units of A between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents overflow when calculating the count parameter to .offset() regardless of the starting point due to past offsets.

• The product of non-zero axis lengths must not exceed isize::MAX.

• Strides must be non-negative.

This function can use debug assertions to check some of these requirements, but it’s not a complete check.

Return the array’s data as a slice, if it is contiguous and in standard order. Return None otherwise.

Note that while the method is similar to ArrayRef::as_slice(), this method transfers the view’s lifetime to the slice, so it is a bit more powerful.

Return the array’s data as a slice, if it is contiguous. Return None otherwise.

Note that while the method is similar to ArrayRef::as_slice_memory_order(), this method transfers the view’s lifetime to the slice, so it is a bit more powerful.

Consume the view and return a reference to the single element in the array.

The lifetime of the returned reference matches the lifetime of the data the array view was pointing to.

use ndarray::{arr0, Array0};

// `Foo` doesn't implement `Clone`.
#[derive(Debug, Eq, PartialEq)]
struct Foo;

let array: Array0<Foo> = arr0(Foo);
let view = array.view();
let scalar: &Foo = view.into_scalar();
assert_eq!(scalar, &Foo);

Consume the mutable view and return a mutable reference to the single element in the array.

The lifetime of the returned reference matches the lifetime of the data the array view was pointing to.

use ndarray::{arr0, Array0};

let mut array: Array0<f64> = arr0(5.);
let view = array.view_mut();
let scalar = view.into_scalar();
*scalar = 7.;
assert_eq!(scalar, &7.);
assert_eq!(array[()], 7.);

Return the array’s data as a slice, if it is contiguous and in standard order. Return None otherwise.

Note that while this is similar to ArrayBase::as_slice_mut(), this method transfers the view’s lifetime to the slice.

Return the array’s data as a slice, if it is contiguous. Return None otherwise.

Note that while this is similar to ArrayBase::as_slice_memory_order_mut(), this method transfers the view’s lifetime to the slice.

Return a shared view of the array with elements as if they were embedded in cells.

The cell view itself can be copied and accessed without exclusivity.

The view acts “as if” the elements are temporarily in cells, and elements can be changed through shared references using the regular cell methods.

Convert into an outer iterator for this view.

Unlike ArrayRef::outer_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an indexed iterator.

Unlike ArrayRef::indexed_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an iterator over an axis.

Unlike ArrayRef::axis_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an iterator over an axis by chunks.

Unlike ArrayRef::axis_chunks_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an outer iterator for this view.

Unlike ArrayRef::outer_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an indexed iterator.

Unlike ArrayRef::indexed_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an iterator over an axis.

Unlike ArrayRef::axis_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an iterator over an axis by chunks.

Unlike ArrayRef::axis_chunks_iter, this methods preserves the lifetime of the data, not the view itself.

Convert into an outer iterator for this view.

Unlike ArrayRef::outer_iter_mut, this methods preserves the lifetime of the data, not the view itself.

Convert into an indexed iterator.

Unlike ArrayRef::indexed_iter_mut, this methods preserves the lifetime of the data, not the view itself.

Convert into an iterator over an axis.

Unlike ArrayRef::axis_iter_mut, this methods preserves the lifetime of the data, not the view itself.

Convert into an iterator over an axis by chunks.

Unlike ArrayRef::axis_chunks_iter_mut, this methods preserves the lifetime of the data, not the view itself.

Split the array view along axis and return one view strictly before the split and one view after the split.

Panics if axis or index is out of bounds.

Examples:

let a = aview2(&[[0, 1, 2, 3],
                 [4, 5, 6, 7],
                 [8, 9, 0, 1]]);

The array view a has two axes and shape 3 × 4:

         ──▶ Axis(1)
        ┌─────┬─────┬─────┬─────┐ 0
      │ │ a₀₀ │ a₀₁ │ a₀₂ │ a₀₃ │
      ▼ ├─────┼─────┼─────┼─────┤ 1
 Axis(0)│ a₁₀ │ a₁₁ │ a₁₂ │ a₁₃ │
        ├─────┼─────┼─────┼─────┤ 2
        │ a₂₀ │ a₂₁ │ a₂₂ │ a₂₃ │
        └─────┴─────┴─────┴─────┘ 3 ↑
        0     1     2     3     4 ← possible split_at indices.

Row indices increase along Axis(0), and column indices increase along Axis(1). Note that we split “before” an element index, and that both 0 and the endpoint are valid split indices.

Example 1: Split a along the first axis, in this case the rows, at index 2.
This produces views v1 and v2 of shapes 2 × 4 and 1 × 4:

let (v1, v2) = a.split_at(Axis(0), 2);

        ┌─────┬─────┬─────┬─────┐       0  ↓ indices
        │ a₀₀ │ a₀₁ │ a₀₂ │ a₀₃ │            along Axis(0)
        ├─────┼─────┼─────┼─────┤ v1    1
        │ a₁₀ │ a₁₁ │ a₁₂ │ a₁₃ │
        └─────┴─────┴─────┴─────┘
        ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄       2
        ┌─────┬─────┬─────┬─────┐
        │ a₂₀ │ a₂₁ │ a₂₂ │ a₂₃ │ v2
        └─────┴─────┴─────┴─────┘       3

Example 2: Split a along the second axis, in this case the columns, at index 2.
This produces views u1 and u2 of shapes 3 × 2 and 3 × 2:

let (u1, u2) = a.split_at(Axis(1), 2);

             u1             u2
        ┌─────┬─────┐┊┌─────┬─────┐
        │ a₀₀ │ a₀₁ │┊│ a₀₂ │ a₀₃ │
        ├─────┼─────┤┊├─────┼─────┤
        │ a₁₀ │ a₁₁ │┊│ a₁₂ │ a₁₃ │
        ├─────┼─────┤┊├─────┼─────┤
        │ a₂₀ │ a₂₁ │┊│ a₂₂ │ a₂₃ │
        └─────┴─────┘┊└─────┴─────┘
        0     1      2      3     4  indices →
                                     along Axis(1)

Splits the view into views of the real and imaginary components of the elements.

use ndarray::prelude::*;
use num_complex::{Complex, Complex64};

let arr = array![
    [Complex64::new(1., 2.), Complex64::new(3., 4.)],
    [Complex64::new(5., 6.), Complex64::new(7., 8.)],
    [Complex64::new(9., 10.), Complex64::new(11., 12.)],
];
let Complex { re, im } = arr.view().split_complex();
assert_eq!(re, array![[1., 3.], [5., 7.], [9., 11.]]);
assert_eq!(im, array![[2., 4.], [6., 8.], [10., 12.]]);

Split the array view along axis and return one mutable view strictly before the split and one mutable view after the split.

Panics if axis or index is out of bounds.

Split the view into multiple disjoint slices.

This is similar to .multi_slice_mut(), but .multi_slice_move() consumes self and produces views with lifetimes matching that of self.

See Slicing for full documentation. See also MultiSliceArg, s!, SliceArg, and SliceInfo.

Panics if any of the following occur:

• if any of the views would intersect (i.e. if any element would appear in multiple slices)
• if an index is out of bounds or step size is zero
• if D is IxDyn and info does not match the number of array axes

Splits the view into views of the real and imaginary components of the elements.

use ndarray::prelude::*;
use num_complex::{Complex, Complex64};

let mut arr = array![
    [Complex64::new(1., 2.), Complex64::new(3., 4.)],
    [Complex64::new(5., 6.), Complex64::new(7., 8.)],
    [Complex64::new(9., 10.), Complex64::new(11., 12.)],
];

let Complex { mut re, mut im } = arr.view_mut().split_complex();
assert_eq!(re, array![[1., 3.], [5., 7.], [9., 11.]]);
assert_eq!(im, array![[2., 4.], [6., 8.], [10., 12.]]);

re[[0, 1]] = 13.;
im[[2, 0]] = 14.;

assert_eq!(arr[[0, 1]], Complex64::new(13., 4.));
assert_eq!(arr[[2, 0]], Complex64::new(9., 14.));

Create an RawArrayView<A, D> from shape information and a raw pointer to the elements.

§Safety

The caller is responsible for ensuring all of the following:

• ptr must be non-null, and it must be safe to .offset() ptr by zero.

• It must be safe to .offset() the pointer repeatedly along all axes and calculate the counts for the .offset() calls without overflow, even if the array is empty or the elements are zero-sized.

  In other words,

  • All possible pointers generated by moving along all axes must be in bounds or one byte past the end of a single allocation with element type A. The only exceptions are if the array is empty or the element type is zero-sized. In these cases, ptr may be dangling, but it must still be safe to .offset() the pointer along the axes.

  • The offset in units of bytes between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents the computed offset, in bytes, from overflowing isize regardless of the starting point due to past offsets.

  • The offset in units of A between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents overflow when calculating the count parameter to .offset() regardless of the starting point due to past offsets.

• The product of non-zero axis lengths must not exceed isize::MAX.

• Strides must be non-negative.

This function can use debug assertions to check some of these requirements, but it’s not a complete check.

Converts to a read-only view of the array.

§Safety

From a safety standpoint, this is equivalent to dereferencing a raw pointer for every element in the array. You must ensure that all of the data is valid, ensure that the pointer is aligned, and choose the correct lifetime.

Split the array view along axis and return one array pointer strictly before the split and one array pointer after the split.

Panics if axis or index is out of bounds.

Cast the raw pointer of the raw array view to a different type

Panics if element size is not compatible.

Lack of panic does not imply it is a valid cast. The cast works the same way as regular raw pointer casts.

While this method is safe, for the same reason as regular raw pointer casts are safe, access through the produced raw view is only possible in an unsafe block or function.

Create an RawArrayViewMut<A, D> from shape information and a raw pointer to the elements.

§Safety

The caller is responsible for ensuring all of the following:

• ptr must be non-null, and it must be safe to .offset() ptr by zero.

• It must be safe to .offset() the pointer repeatedly along all axes and calculate the counts for the .offset() calls without overflow, even if the array is empty or the elements are zero-sized.

  In other words,

  • All possible pointers generated by moving along all axes must be in bounds or one byte past the end of a single allocation with element type A. The only exceptions are if the array is empty or the element type is zero-sized. In these cases, ptr may be dangling, but it must still be safe to .offset() the pointer along the axes.

  • The offset in units of bytes between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents the computed offset, in bytes, from overflowing isize regardless of the starting point due to past offsets.

  • The offset in units of A between the least address and greatest address by moving along all axes must not exceed isize::MAX. This constraint prevents overflow when calculating the count parameter to .offset() regardless of the starting point due to past offsets.

• The product of non-zero axis lengths must not exceed isize::MAX.

• Strides must be non-negative.

This function can use debug assertions to check some of these requirements, but it’s not a complete check.

Converts to a read-only view of the array.

§Safety

From a safety standpoint, this is equivalent to dereferencing a raw pointer for every element in the array. You must ensure that all of the data is valid, ensure that the pointer is aligned, and choose the correct lifetime.

Converts to a mutable view of the array.

§Safety

From a safety standpoint, this is equivalent to dereferencing a raw pointer for every element in the array. You must ensure that all of the data is valid, ensure that the pointer is aligned, and choose the correct lifetime.

Split the array view along axis and return one array pointer strictly before the split and one array pointer after the split.

Panics if axis or index is out of bounds.

Cast the raw pointer of the raw array view to a different type

Panics if element size is not compatible.

Lack of panic does not imply it is a valid cast. The cast works the same way as regular raw pointer casts.

While this method is safe, for the same reason as regular raw pointer casts are safe, access through the produced raw view is only possible in an unsafe block or function.