Hands-On Python Deep Learning for the Web

Tensors

Like TensorFlow, the central data processing unit in TF.js is tensors. Goodfellow et al. (in their book on deep learning) make the following observation:

In the general case, an array of numbers arranged on a regular grid with a variable number of axes is known as a tensor.

Simply described, a tensor is a container of one- or multi-dimensional arrays. The following are some examples of tensors that you may already know:

• Scalar (a rank zero tensor)
• Vector (a one-dimensional or rank-one tensor)
• Matrix (a two-dimensional or rank-two tensor)

We can create a tensor with respect to a given shape in TF.js as shown:

const shape = [2, 3]; // 2 rows, 3 columnsconst a = tf.tensor([4.0, 2.0, 5.0, 15.0, 19.0, 27.0], shape);

a is a tensor that ...