LMS Offline

A non-iterative method for computing a weight matrix for associating input vectors and target vectors [link to training set] in a training set.

Some practical tips

Wiener-Hopf

Moore-Penrose.

Moore-Penrose is stable in some cases when Winer Hopf is not.

Ridge regression and adding noise help when a matrix that is rank deficient. Prevents over-fitting, makes rank slightly higher.

You have to make sure the output node range is greater than the range of the target values. 

Sometimes it helps to change output nodes to linear.

Parameters

Play: Trains the network until told to stop.

Iterate: Iterates the network once.

Train: Trains the network in batches of the size given by the Batch parameter.

References

http://www.scholarpedia.org/article/Echo_state_network