Was in a neural network predictive regression feel its principle and traditional nonlinear regression, don't know what is the difference between them, some people say that a need theory knowledge, or we'll try one by one, but the number of hidden layer nodes of neural network activation function is not each one tries to build a model, too, in addition to the neural network may not be a dependent variable, I couldn't figure out what's the difference between temporarily, can have a big light for me

CodePudding user response:

First of all, I let's look at the traditional nonlinear regression method:
1, can be converted into linear regression problems: take logarithm etc, and then according to the least squares method
2, and cannot be converted to linear regression: one is based on professional knowledge, or is derived theoretically by experience that the second is in professional knowledge helpless situation, through mapping and observation a scatter diagram to determine the curve type,
The second category, a problem we often is relatively weak,
And neural network method is the problem for the digital abstract characteristics, namely on the network weights and bias, although at the time of solving numerical characteristics also use the similar to the least squares method, but in fact is the use of computer performance, the nonlinear problem into a general available least-square method to solve linear problems