the learning rate inportance

Hello, Plz, what is the importance of "learning rate" how can we comput it, and st it via NNtoolbox? thank you, BR

[Jan]

Moar details, moar gud nswr.

Answer 1

[Greg Heath]

I have designed zillions of NNs with MATLAB and do not remember ever having to use a nondefault value of lr.

Typically, the only default overrides I have ever had to use for regression (FITNET) and classification (PATTERNNET) were

 1. Initial state of the RNG
 2. Number of hidden nodes

For timeseries I 've had to also override default

 3. Input and feedback delays (based on significant delays of 
    the target autocorrelation function and target-input 
    crosscorrelation function)
 4. datadivision type
 5. datadivision fractions

Hope this helps.

Thank you for formally accepting my answer

Greg