CtrlK
A shape in the context of machine learning and deep learning refers to the structure or dimensionality of a tensor, which is a multi-dimensional array of numerical values. Tensors are the fundamental building blocks of many machine learning models and frameworks, such as TensorFlow and PyTorch. The shape of a tensor is characterized by the number of dimensions it has, known as its rank, and the size of each dimension, known as its axes.
The rank of a tensor is the number of dimensions it possesses. Scalars, which are single numerical values, have a rank of 0. Vectors, which are one-dimensional arrays, have a rank of 1. Matrices, which are two-dimensional arrays, have a rank of 2. Higher-dimensional arrays are called tensors, and their rank is equal to the number of dimensions they have.
The axes of a tensor are the individual dimensions along which the data is organized. The size of each axis is the number of elements present in that specific dimension. For example, a 3x3 matrix has two axes: one of size 3 in the row direction and another of size 3 in the column direction.
In many machine learning operations, it is necessary to reshape or manipulate the shape of tensors to match the input or output requirements of different layers or algorithms. Common operations include:
Imagine you have a big box of building blocks. Each block can be thought of as a number. A shape tells us how the blocks are arranged inside the box. For example, if the shape is `(2, 3)`, that means there are 2 rows and 3 columns of blocks. The word tensor is just a fancy name for a box of blocks with a certain shape. In machine learning, we use these tensors to store and process data, and sometimes we need to change their shape to make them fit together nicely.