I have questions about adapt function of neural network toolbox

I use NNtoolbox to dynamic modeling.

I have 7 datasets. So I want to use 3 datasets to train and use 4 dataset to test.

And I will use adapt function.

By the way.

Where can I find more specific description about adapt algorithm than help of matlab program?

Or can you tell me papers about adapt algorithm?

Answer 1

[Greg Heath]

The basic differences between adapt and train are:

1. Train uses batch updating. Each epoch all of the data is passed through the model with the current weights. Errors are accumulated and all of the weights updates are made at once at the end of the epoch. Since this is amenable to matrix algebra, batch updating tends to be relatively fast for large data sets.

2. Adapt uses sequential or online updating. Data cases are presented individually. After each is case is presented, it's error is determined and the weights are updated accordingly. Therefore, there are N updates per epoch instead of one. After several epochs, each update is performed on a slightly improved model.

Hope this helps.

Greg

[Greg Heath]

See the Toolbox documentation.

doc

Neural Network Toolbox

-Users' Guide

--Network Objects, Data and Training Styles

---Training Styles (Adapt and Train)

Also use the DOC and TYPE commands in addition to HELP.

doc adapt

doc train

type adapt

type train

[Greg Heath]

Thank you for accepting my answer!