Tensor

Tensor expresses the data. Scalar, Vector, Matrix, and more multidimensional array are supported.

Basic usage

To create a tensor from a JS number or array, use mt.tensor() API.

const scalar = hk.tensor(0)
const vector = hk.tensor([1, 2, 3])
const matrix = hk.tensor([
  [1, 2, 3],
  [4, 5, 6],
])

To get a JS value, use tensor.toArray() method. It's async process.

const vector = hk.tensor([1, 2, 3])

const arr = await vector.toArray() // [1, 2, 3]

Calculation

Of course, you can calculate these tensors. Some operators are supported.

Addition

Use tensor.add(other) method.

const a = hk.tensor(1)
const b = hk.tensor(2)

const added = a.add(b) // 3

Tensor states

There is three kinds of tensor states: Creating, Calculating, and Resolved.

CreatingTensor

CreatingTensor means tensors that may or may not have been created.

For example, tensor is that code is CreatingTensor.

const tensor = hk.tensor(1)

Internally, there is a possibility that tensor created, there is also a possibility that tensor hasn't created yet.

CalculatingTensor

CalculatingTensor means tensors which don't have the result. It means it hasn't calculated yet. It is created when you use calculate methods such as add(), sub(), and others.

const creating = hk.tensor(1)
const calculating = creating.add(hk.tensor(2))

In that code, calculating means 3 but the result hasn't computed.

ResolvedTensor

ResolvedTensor is the tensor the result was confirmed. You can change state to ResolvedTensor by using await.

An example resolving CreatingTensor:

const creating = hk.tensor(1)
const resolved = await creating // Resolved!

Or, you can resolve CalculatingTensor:

const calculating = hk.tensor(1).add(hk.tensor(1))
const resolved = await calculating // Resolved!