# MLShapedArray A machine learning collection type that stores scalar values in a multidimensional array. ``` struct MLShapedArray where Scalar : MLShapedArrayScalar ``` ## Overview A shaped array is a multidimensional array type that’s the Swift counterpart to [`MLMultiArray`](/documentation/CoreML/MLMultiArray). [`MLShapedArray`](/documentation/CoreML/MLShapedArray) is one of the underlying types of `MLFeatureValue` that stores scalar values. You can convert a shaped array to an [`MLMultiArray`](/documentation/CoreML/MLMultiArray) with its [`init(_:)`](/documentation/CoreML/MLMultiArray/init(_:)-wk41) initializer, and convert back to a shaped array with its [`init(_:)`](/documentation/CoreML/MLShapedArray/init(_:)) initializer. All elements in an [`MLShapedArray`](/documentation/CoreML/MLShapedArray) are of the same type, and that type must conform to [`MLShapedArrayScalar`](/documentation/CoreML/MLShapedArrayScalar): - - - Each dimension in a shaped array is typically significant or meaningful. For example, a model could have an input that accepts images as a three-dimensional array of pixels, C x H x W. The first dimension, *C*,_ _represents the number of color channels, and the second and third dimensions, *H* and *W*, represent the image’s height and width, respectively. The number of dimensions and size of each dimension define the shaped array’s *shape*. > Note: > Some models use a one-dimensional multiarray for an input or output. This type of shaped array is conceptually identical to a conventional . A shaped array’s [`shape`](/documentation/CoreML/MLMultiArray/shape) property is an integer array in which each element defines the size of the corresponding dimension. To inspect the shape and constraints of a model’s multiarray input or output feature: 1. Access the model’s [`modelDescription`](/documentation/CoreML/MLModel/modelDescription) property. 2. Find the multiarray input or output feature in the model description’s [`inputDescriptionsByName`](/documentation/CoreML/MLModelDescription/inputDescriptionsByName) or [`outputDescriptionsByName`](/documentation/CoreML/MLModelDescription/outputDescriptionsByName) property, respectively. 3. Access the feature description’s [`multiArrayConstraint`](/documentation/CoreML/MLFeatureDescription/multiArrayConstraint) property. 4. Inspect the multiarray constraint’s [`shape`](/documentation/CoreML/MLMultiArrayConstraint/shape) and [`shapeConstraint`](/documentation/CoreML/MLMultiArrayConstraint/shapeConstraint). ## Topics ### Creating a shaped array [`init(scalar:)`](/documentation/CoreML/MLShapedArray/init(scalar:)) Creates a shaped array with exactly one value and zero dimensions. [`init(scalars:shape:)`](/documentation/CoreML/MLShapedArray/init(scalars:shape:)) Initialize with a sequence and the shape. [`init(mutating:shape:)`](/documentation/CoreML/MLShapedArray/init(mutating:shape:)) Creates a new `MLShapedArray` using a pixel buffer as the backing storage. ### Creating a shaped array from another type [`init(_:)`](/documentation/CoreML/MLShapedArray/init(_:)) [`init(concatenating:alongAxis:)`](/documentation/CoreML/MLShapedArray/init(concatenating:alongAxis:)) Merges a sequence of shaped arrays into one shaped array along an axis. ### Creating a shaped array with pointers to memory [`init(unsafeUninitializedShape:initializingWith:)`](/documentation/CoreML/MLShapedArray/init(unsafeUninitializedShape:initializingWith:)) Creates a shaped array from a shape and a closure that initializes its memory. ### Creating a shaped array from data [`init(data:shape:)`](/documentation/CoreML/MLShapedArray/init(data:shape:)) Creates a shaped array from a block of data and a shape. [`init(data:shape:strides:)`](/documentation/CoreML/MLShapedArray/init(data:shape:strides:)) Creates a shaped array from a block of data, a shape, and strides. ### Shaping the array [`changingLayout(to:)`](/documentation/CoreML/MLShapedArray/changingLayout(to:)) Returns a copy with the specified buffer layout. [`expandingShape(at:)`](/documentation/CoreML/MLShapedArray/expandingShape(at:)) Returns a new shaped array with expanded dimensions. [`reshaped(to:)`](/documentation/CoreML/MLShapedArray/reshaped(to:)) Returns a new reshaped shaped array. [`squeezingShape()`](/documentation/CoreML/MLShapedArray/squeezingShape()) Returns a new squeezed shaped array. [`transposed()`](/documentation/CoreML/MLShapedArray/transposed()) Returns a new transposed shaped array. [`transposed(permutation:)`](/documentation/CoreML/MLShapedArray/transposed(permutation:)) Returns a transposed shaped array using a custom permutation. ### Reading and writing the pixel buffer [`withMutablePixelBufferIfAvailable(_:)`](/documentation/CoreML/MLShapedArray/withMutablePixelBufferIfAvailable(_:)) Writes to the underlying pixel buffer. [`withPixelBufferIfAvailable(_:)`](/documentation/CoreML/MLShapedArray/withPixelBufferIfAvailable(_:)) Reads the underlying pixel buffer. ### Modifying a shaped array [`withUnsafeMutableShapedBufferPointer(using:_:)`](/documentation/CoreML/MLShapedArray/withUnsafeMutableShapedBufferPointer(using:_:)) Calls the given closure with a pointer to the array’s mutable storage that has a specified buffer layout. ### Encoding and decoding [`init(from:)`](/documentation/CoreML/MLShapedArray/init(from:)) Creates a shaped array from a decoder. [`encode(to:)`](/documentation/CoreML/MLShapedArray/encode(to:)) Encode a shaped array. --- Copyright © 2026 Apple Inc. All rights reserved. | [Terms of Use](https://www.apple.com/legal/internet-services/terms/site.html) | [Privacy Policy](https://www.apple.com/privacy/privacy-policy)