DTensor

interface DTensor : Differentiable<DTensor>

Interface for a differentiable tensor.

DTensor is an interface that is implemented as either

a FloatTensor, which is a wrapper around an array of floats, or
a ForwardTensor for forward differentiation, or
a ReverseTensor for reverse mode differentiation.

Functions

get

open operator fun get(index: Int): DTensor

open operator fun get(vararg indices: Int): DTensor

toCodeString

open fun toCodeString(): String

wrap

open override fun wrap(wrapper: Wrapper): DTensor

Wrapper around the tensor

Properties

derivativeID

abstract val derivativeID: DerivativeID

Each derivative is assigned a unique DerivativeID

indices

open val indices: Iterator<IntArray>

An iterator over the indices for the tensor

isScalar

open val isScalar: Boolean

True if the tensor is a scalar.

operations

abstract val operations: Operations

The operations available on a tensor.

primal

abstract val primal: DTensor

primal points to the actual tensor

rank

open val rank: Int

The number of dimensions in the tensor's shape. rank 0 - Scalar rank 1 - 1D array or 1D tensor rank 2 - 2D matrix or 2D tensor rank 3 - 3D tensor ... rank N - ND tensor

shape

open val shape: Shape

shape indicates the number of dimension of a tensor and the length of each dimension. If the shape of the tensor is 3x4x5 then the value of shape is Shape(3,4,5).

size

open val size: Int

The total number of elements of this tensor.

Inheritors

DScalar

FloatTensor

ForwardTensor

ReverseTensor

RandomTensorWrapper

TracingTensor

Extensions

allAxes

val DTensor.allAxes: IntArray

basePrimal

fun DTensor.basePrimal(): FloatTensor

Returns the primal value as a FloatTensor.

broadcastToShape

fun DTensor.broadcastToShape(shape: Shape): DTensor

concat

fun DTensor.concat(right: DTensor, axis: Int): DTensor

Concatenate two tensors along the provided axis.

fun DTensor.concat(right: DTensor): DTensor

div

operator fun DTensor.div(right: DScalar): DTensor

operator fun DTensor.div(right: Float): DTensor

operator fun DTensor.div(that: DTensor): DTensor

infix fun DTensor.eq(other: DTensor): DTensor

infix fun DTensor.eq(other: Float): DTensor

expand

fun DTensor.expand(newShape: Shape): DTensor

expandToTangent

fun DTensor.expandToTangent(tangent: DTensor): DTensor

flatten

fun DTensor.flatten(): DTensor

Flattens a tensor into a vector (a tensor with rank 1).

fun DTensor.flatten(startDim: Int = 0, endDim: Int = rank - 1): DTensor

Returns a tensor with dimensions startDim, endDim (inclusive range) flattened.

flip

fun DTensor.flip(axes: IntArray): DTensor

@JvmName(name = "ExtensionVarargFlip")

fun DTensor.flip(vararg axes: Int): DTensor

Create a tensor of the same shape, with all elements flipped across the specified axes

gather

fun DTensor.gather(indices: List<Int>, axis: Int, paddingIndex: Int = -1): DTensor

Return a tensor made up of slices taken from the input tensor at the given indices along the given axis.

infix fun DTensor.ge(other: DTensor): DTensor

infix fun DTensor.ge(other: Float): DTensor

infix fun DTensor.gt(other: DTensor): DTensor

infix fun DTensor.gt(other: Float): DTensor

innerProduct

fun DTensor.innerProduct(b: Shape, right: DTensor): DTensor

Takes a tensor of shape T, a shape B, and a tensor of shape T and returns a tensor of shape T which is the inner product of the two.

into

infix fun <T : Layer<T>> DTensor.into(layer: Layer<T>): DTensor

infix fun DTensor.le(other: DTensor): DTensor

infix fun DTensor.le(other: Float): DTensor

leftTranspose

fun DTensor.leftTranspose(a: Shape, b: Shape): DTensor

Given the shape A,B,D (where A, B and D are lists of Ints, and D is possibly empty), this function converts a tensor of shape A,B,D to a tensor of shape B,A,D, shuffling the data so that the element at position i,j,k is at position j,i,k (where i, j and k are lists of integers corresponding to the shape A, B, and D).

logSoftmax

fun DTensor.logSoftmax(axis: Int): DTensor

infix fun DTensor.lt(other: DTensor): DTensor

infix fun DTensor.lt(other: Float): DTensor

matdiv

fun DTensor.matdiv(that: DTensor): DTensor

matmul

fun DTensor.matmul(right: DTensor): DTensor

Matrix multiply of two tensors. See https://pytorch.org/docs/stable/generated/torch.matmul.html for the specification.

fun DTensor.matmul(right: DTensor, a: Shape, b: Shape, c: Shape, d: Shape): DTensor

Generalized matrix multiply of two tensors.

minus

operator fun DTensor.minus(right: DTensor): DTensor

Tensor subtraction.

operator fun DTensor.minus(right: Float): DTensor

momentumUpdated

fun DTensor.momentumUpdated(new: DTensor, momentum: Float): DTensor

Returns the current (this) tensor updated by the new tensor (new) scaled by momentum

infix fun DTensor.ne(other: DTensor): DTensor

infix fun DTensor.ne(other: Float): DTensor

outerProduct

infix fun DTensor.outerProduct(right: DTensor): DTensor

plus

operator fun DTensor.plus(right: DTensor): DTensor

Tensor addition.

operator fun DTensor.plus(right: Float): DTensor

pow

fun DTensor.pow(x: DScalar): DTensor

fun DTensor.pow(x: Int): DTensor

fun DTensor.pow(exponent: DTensor): DTensor

fun DTensor.pow(exponent: Float): DTensor

primal

fun DTensor.primal(derivativeID: DerivativeID): DTensor

relu

@JvmName(name = "DTensorRelu")

fun DTensor.relu(): DTensor

reshape

fun DTensor.reshape(vararg newShape: Int): DTensor

fun DTensor.reshape(newShape: Shape): DTensor

rightTranspose

fun DTensor.rightTranspose(a: Shape, b: Shape): DTensor

Given the shape A,B,D (where A, B and D are lists of Ints, and D is possibly empty), this function converts a tensor of shape D,A,B to a tensor of shape D,B,A, shuffling the data so that the element at position i,j,k is at position i,k,j (where i, j and k are lists of integers corresponding to the shape D, B, and A).

scatter

fun DTensor.scatter(indices: List<Int>, axis: Int, newShape: Shape, paddingIndex: Int = -1): DTensor

slice

fun DTensor.slice(start: Int, end: Int, axis: Int = 0): DTensor

Return a slice of the input tensor, which includes only some of the indices at axis. Specifically, it includes indices from start (inclusive) until end (exclusive).

softmax

@JvmName(name = "DTensorSoftmaxExt")

fun DTensor.softmax(axis: Int): DTensor

split

fun DTensor.split(shapes: List<Shape>): List<DTensor>

Takes a tensor, and a list of the desired shapes of the components to split it into, and breaks the values up into a list of DValues. For a desired shape that is an empty List, a DScalar is produced. Otherwise a DTensor is produced with the desired shape.

squeeze

fun DTensor.squeeze(axis: Int): DTensor

Returns a tensor with the singleton dimension at the specified position of the shape removed.

stack

fun DTensor.stack(right: DTensor, axis: Int = 0): DTensor

Stack two tensors along the provided new axis.

stats

fun DTensor.stats(): Pair<DScalar, DScalar>

Compute the mean and variance of the data.

sum

fun DTensor.sum(): DScalar

@JvmName(name = "varargSum")

fun DTensor.sum(vararg axes: Int, keepDims: Boolean = false): DTensor

fun DTensor.sum(axes: IntArray = IntArray(rank) { it }, keepDims: Boolean = false): DTensor

Sum over given axes. If keepDims is true, the original rank of input is preserved. Otherwise, the dimensions provided by axis are removed from the output shape.

times

operator fun DTensor.times(right: Float): DTensor

operator fun DTensor.times(right: DScalar): DTensor

operator fun DTensor.times(that: DTensor): DTensor

transpose

fun DTensor.transpose(axes: IntArray = this.allAxes.reversedArray()): DTensor

Returns x transposed over axes. If no axes are provided, tensor is transposed over all axes (that is, the order of the axes are reversed).

unaryMinus

operator fun DTensor.unaryMinus(): DTensor

unsqueeze

fun DTensor.unsqueeze(axis: Int): DTensor

Returns a new tensor with a dimension of size one inserted at the specified position of its shape.

view

fun DTensor.view(indices: IntArray): DTensor

fun DTensor.view(index: Int, axis: Int): DTensor

fun DTensor.view(index: IntRange, axis: Int): DTensor

withChange

fun DTensor.withChange(index: Int, axis: Int, replacementValue: DTensor): DTensor

fun DTensor.withChange(indices: IntArray, replacementValue: DTensor): DTensor

fun DTensor.withChange(index: IntRange, axis: Int, replacementValue: DTensor): DTensor