1.MatalbToolbox:nnstart/nprtool/nftool/nctool/ntstool
nprtool:pattern recognization
nftool:function fitting
nctool:data clustering
ntstool:time series
2.Steps to solve a problem:
collect data–create the network–configure the network–initialize the weights and biases–train the network–validate the network–use the network
3.Train algorithms:
trainlm:Jocobian calculations.More memory and less time.For most problems.(Levenberg-Marquardt)
trainbr:For noisy,dirty dataset.(Bayesian Regularization)
trainscg:For large dataset.Less memory.(Scaled Conjugate Gradient)
4.validation:
generalization stops improving.
5.The training continued until the validation error failed to decrease for six iterations(validation stop).
6.choose a train algorith and number of neurons in hidden layer,and start train
7.Then u get train result:
8.Plots:
Regression:the relation between targets and outputs.For a perfect fit,data should fall along a 45 degree line.
ErrHist:distribution of error(Targets-Outputs).To discover the outliers as to find out bad data.
9.If u are dissatisfied with the fitting result,u can try:
Retrain.
Increase the network size.
Import larger data set.
10.If u are satisfied with the result,click”next”:
save network/perfomance/error
save as scripts to learn more.