Try using Tensorflow and Numpy while solving your doubts.

Python numpy CNN TensorFlow Tensor

purpose

For years, many people have reported on how to use Tensorflow and Numpy. Referring to them, I will try to follow up while organizing the parts that I wondered about.

For the time being, the purpose of use shall be CNN.

I'm still in the beginning, but the description that I think is somewhat useful is below.

What is a tensor? 　　　　　But what is a tensor?

What is a tensor?

What is the tensor in the name of Tensorflow? If you look it up on the net, you can see that the tensor itself is quite difficult. There are many things that are related to Tensorflow and explain tensor, but I feel that only the explanation of tensors is difficult and does not correspond to Tensorflow.

As for myself, as explained in the following blog, I thought it would be good to understand it as "multidimensional array". (I felt that the following blog succeeded in explaining.) ）

HELLO CYBERNETICS "TensorS You Need to Know Before Starting TensorFlow (Update: To a More General Topic)" https://www.hellocybernetics.tech/entry/2016/12/01/223834

But what is a tensor?

It's been about a year since I first wrote this article、、、、 All of a sudden, I feel like I know what a tensor is. (I mean, I feel like I know what I'm trying to convey on the site I'm trying to explain.) ） I don't know enough to explain it, but in a certain field, it is a reasonably complex and key technology. In that area, I think it's not a sequence of numbers like a "multidimensional array", but something more meaningful. However, in the field of deep learning, at best, I think there is no problem with understanding "multidimensional arrays".

Suddenly, the atmosphere of Tensor was grasped by the site

Source: https://medium.com/@quantumsteinke/whats-the-difference-between-a-matrix-and-a-tensor-4505fbdc576c

A tensor is often thought of as a generalized matrix. That is, it could be a 1-D matrix (a vector is actually such a tensor), a 3-D matrix (something like a cube of numbers), even a 0-D matrix (a single number), or a higher dimensional structure that is harder to visualize. The dimension of the tensor is called its rank. But this description misses the most important property of a tensor! A tensor is a mathematical entity that lives in a structure and interacts with other mathematical entities. If one transforms the other entities in the structure in a regular way, then the tensor must obey a related transformation rule.

Once you get a feel for Tensor, you can understand the wiki description somehow.

Source: https://ja.wikipedia.org/wiki/ tensor

A tensor is a generalization of a linear quantity or linear geometric concept that can be expressed as a multidimensional array if the basis is chosen. However, the tensor itself is a subject that is determined without depending on a particular coordinate system. For each tensor, the number of sets of subscripts of arrays necessary to describe the corresponding quantities is called the rank of the tensor. For example, scalar quantities such as mass and temperature are understood to be tensors of rank 0. Similarly, vectoric quantities such as force and momentum are tensors of rank 1, and linear transformations that represent anisotropic relationships between forces and acceleration vectors are represented by tensors of rank 2. It should be noted that what is often referred to as a "tensor" in physics and engineering is actually a "tensor field," a function that returns a tensor quantity with position and time as arguments. In any case, in order to understand tensor fields, it is essential to understand the concept of tensors themselves.