Questions about neural networks
When you perform logistic regression in neural networks, you specify the learning rate, the number of updates, etc., right?On the other hand, I do not remember specifying such a thing when performing logistic regression in a generalized linear model (glm() in R).
Don't you set the learning rate in the generalized linear model? Or is something done internally?
I would appreciate it if you could show me both formulas and explain them.
python r machine-learning neural-network
It's called a formula, but I thought I could explain why there are no parameters, so I'll just explain.
Logistic regression estimates parameters for a given probability distribution from the data, assuming that certain observational data follow a predetermined probability distribution.Because it can be solved mathematically, the parameters are uniquely determined by providing data.
Neural networks become much more complex formulas, and the best likelihood estimation for all of these inputs is probably impossible to solve.However, for a particular data, you can find the gradient that optimizes the error between the output and the solution of the network as a function.In other words, if you fix the data and the corresponding answers, you have a structure where the gradient method is applied to the error function.For each data, the weight is gradually updated by this gradient method.Repeat this endlessly until the error is minimized.This is the internal logic of what is said to be the reverse propagation of errors in neural networks.And how much this weight is updated is called the learning rate.
Neural networks (or multilayer perceptrons) are often used in pattern matches because they have the property that error reverse propagation can only be accomplished by matrix operations, and because increasing intermediate nodes can provide infinite accuracy.
© 2024 OneMinuteCode. All rights reserved.