What is ONE HOT ENCODING?

An efficient method of encoding the classes to train a network.

One Hot Encoding

[1,0,0]: Class A
[0,1,0]: Class B
[0,0,1]: Class C

Efficient Encoding

0: Class A
1: Class B
2: Class C

In neural networks when we need to pick a class from classes, we have output nodes equal to the number of classes. Each node shows the probability that it may matches Class A, Class B or Class C. It may look like this: Class A has probability equal to 0.1 Class B has probability equal to 0.2 Class C has probability equal to 0.7

Although it looks inefficient from the storage perspective, but it is very efficient for the training. Moreover, it complements argmax function, that saves us writing a lot of code.