Levenberg-Marquardt algorithm alternatives in Neurolab

Question

I am porting neural network code in Matlab that uses Levenberg-Marquardt training algorithm to Python. I am using Neurolab, which I believe does not come with Levenberg-Marquardt training algorithm. Is there any alternative training algorithm I can use? I am using a feed forward multilayer network. Thank you!

Answer 1

This is too long for a comment.

Levenberg-Marquardt is an optimization method for estimating the parameters for a neural network. There are many other algorithms (back propagation, quick prop, and many more). My favorite all text web page is the Neural Network FAQ, which explains many of these techniques.

However, the end result is a trained neural network. So, you can try other training techniques available in Neurolab (which I have not used). Some may perform better than others, but they are all doing the same thing -- setting appropriate parameter values.

Answer 2

A new (recurrent) neural network toolbox called pyrenn has been released, which exploits the Levenberg-Marquardt algorithm. I have actually used it myself very recently and got same results as in Matlab.

You can install it using pip.