Hands-On Machine Learning with TensorFlow.js

Tensor construction

A tensor is a data structure that holds metadata such as shapes, data types, and pointers for the underlying buffer. To construct a tensor, we need to provide some data as parameters. tf.tensor allows us to create the arbitrary shape of the tensor:

tf.tensor([1, 2, 3, 4], [2, 2]).print();    Tensor    [[1, 2],     [3, 4]]

The API recognizes the dimensions of the given JavaScript array, which allows us to specify the shape explicitly:

tf.tensor([[1, 2], [3, 4]]).print();    Tensor    [[1, 2],     [3, 4]]

There are also several rank-specific APIs that we can use to create tensors:

• tf.tensor1d: Tensor1D
• tf.tensor2d: Tensor2D
• tf.tensor3d: Tensor3D
• tf.tensor4d: Tensor4D
• tf.tensor5d: Tensor5D
• tf.tensor6d: Tensor6D
• tf.scalar: Scalar

These APIs ...