CV review: softmax code implementation

softmax function, also known as normalized exponential function.It is a binary classification function sigmoid in multi-classThe purpose of the generalization is to display the results of multi-classification in the form of probability.

1) Convert the forecast to a non-negative number

    Convert a value to an exponential function

2) The sum of the probabilities of various prediction results is equal to 1

     To ensure that the sum of the probabilities of each prediction result equals 1.We just need to normalize the transformed result.This is done by dividing the converted result by the sum of all the converted results, which can be understood as a percentage of the total converted results.This gives an approximate probability.

So the final formula of softmax is:

Code implementation:

Because the magnification of the exponential function is too obvious, if directly use the softmax calculation formulato implement the function, it will easily lead to data overflow (overflow).Therefore, we use its properties when implementing the function: first process the input data, and then use the calculation formula to calculate.Specifically, the implementation steps are as follows